Disinformation in Democracies: Strengthening Digital Resilience in Latin America

Home , Electoral fraud, Joice Hasselmann, Laura Chinchilla, Superior Electoral Court

Atlantic Council ADRIENNE ARSHT LATIN AMERICA CENTER

Disinformation Strengthening in Democracies:Digital Resilience in Latin America The Adrienne Arsht Latin America Center broadens understanding of regional transformations through high-impact work that shapes the conversation among policymakers, the business community, and civil society. The Center focuses on Latin America’s strategic role in a global context with a priority on pressing political, economic, and social issues that will define the trajectory of the region now and in the years ahead. Select lines of programming include: Venezuela’s crisis; Mexico-US and global ties; China in Latin America; Colombia’s future; a changing Brazil; Central America’s trajectory; combatting disinformation; shifting trade patterns; and leveraging energy resources. Jason Marczak serves as Center Director.

The Atlantic Council’s Digital Forensic Research Lab (DFRLab) is at the forefront of open-source reporting and tracking events related to security, democracy, technology, and where each intersect as they occur. A new model of expertise adapted for impact and real-world results, coupled with efforts to build a global community of #DigitalSherlocks and teach public skills to identify and expose attempts to pollute the information space, DFRLab has operationalized the study of disinformation to forge digital resilience as humans are more connected than at any point in history.

For more information, please visit www.AtlanticCouncil.org.

This report is written and published in accordance with the Atlantic Council Policy on Intellectual Independence. The authors are solely responsible for its analysis and recommendations. The Atlantic Council and its donors do not determine, nor do they necessarily endorse or advocate for, any of this report’s conclusions.

© 2019 The Atlantic Council of the United States. All rights reserved. No part of this publication may be reproduced or transmitted in any form or by any means without permission in writing from the Atlantic Council, except in the case of brief quotations in news articles, critical articles, or reviews. Please direct inquiries to:

Atlantic Council 1030 15th Street NW, 12th Floor Washington, DC 20005

ISBN-13: 978-1-61977-524-4

March 2019 Disinformation in Democracies: Strengthening Digital Resilience in Latin America

This report is a collaborative effort between the Digital Forensic Research Lab (DFRLab) and the Adrienne Arsht Latin America Center of the Atlantic Council.

By: Luiza Bandeira, Donara Barojan, Roberta Braga, Jose Luis Peñarredonda, Maria Fernanda Pérez Argüello disinformation in democracies Foreword

s the sun rose on 2019, democracies Governments, technology and social media com- across Latin America awoke to new lead- panies, civil society, and media don’t yet fully under- ers and new administrations looking to stand the challenges disinformation poses to democ- fulfill promises of change. In Brazil, the racy. These stakeholders, each with different sets of Anew year marked Jair Bolsonaro’s inauguration, and in motivations, do not yet fully see eye to eye on the best Mexico and Colombia, Andrés Manuel López Obrador way to address the rise of influence operations online. and Iván Duque, respectively, continued to implement Meanwhile, regulatory frameworks remain outdated as their nascent government agendas. technologies evolve at a faster pace than the laws that Having recently led the Organization of American regulate online abuses and punish bad actors. States (OAS) observation missions of some of these How can society protect freedom of expression elections and as these country agendas were taking while simultaneously combatting disinformation in the shape, I laid out my own vision for what democracy digital realm? And how can public and private-sec- holds in the short and long terms—particularly the tor stakeholders work alongside media and others to digital dangers that are likely to affect political con- effectively manage or contain threats brought on by tests and their legitimacy. disinformation and misinformation? Digital technologies have immense potential This report seeks to contribute to ongoing discus- to improve lives. But, as the 2018 elections in Latin sions with proposals for how to address these growing America showed, they can also be used to exacerbate challenges. These conversations must be had in Latin the spread of disinformation and misinformation in America, but the implications and lessons extend far ways that can impact political decision-making and beyond the region. Now more than ever, comprehen- electoral outcomes. sive solutions are needed to mitigate the negative In 2018, the Atlantic Council’s Digital Forensic impact of disinformation while promoting the many Research Lab (DFRLab) and Adrienne Arsht Latin benefits of direct citizen engagement. America Center partnered to strengthen digital This year, more than eighty elections are taking resilience in elections by identifying, exposing, and place worldwide. Many will decide leaders of key explaining instances of disinformation, working along- democracies—in Latin America, Europe, Asia, and side Latin American media, civil society, and aca- beyond. Failing to create deeper understanding, and demic organizations. The effort, #ElectionWatch Latin to better align efforts to address the spread of false America, was—and continues to be—a necessary step information around elections, could significantly in addressing the spread of disinformation and misin- undermine public trust in the democratic process formation facilitated by the rapid permeation of online for decades. But, getting it right will bring immense technologies. society-wide benefits.

Laura Chinchilla Chair, Kofi Annan Foundation; Former President of Costa Rica; Member, Adrienne Arsht Latin America Center Advisory Council, Atlantic Council

1 disinformation in democracies

Table of Contents

EXECUTIVE SUMMARY �� 3

INTRODUCTION �� 4

DISINFORMATION IN THE 2018 ELECTIONS

BRAZIL �� 6

MEXICO �� 20

COLOMBIA �� 3O

RECOMMENDATIONS �� 42

CONCLUSION �� 46

ACKNOWLEDGMENTS �� 47

GLOSSARY OF KEY TERMS �� 48

ABOUT THE AUTHORS �� 50

ENDNOTES �� 52

2 Executive Summary

etween May and October 2018, three of In Brazil, Atlantic Council research conducted in real Latin America’s largest democracies— time found that disinformation comprised primarily Brazil, Mexico, and Colombia—elected new organic disinformation—driven by polarization and a heads of state, congress members, and lack of trust in institutions. In Colombia, the Atlantic Blocal officials in a series of polarized, contentious elec- Council observed a similar trend, exacerbated at tions. In a region in which the number of cell phones times by political leaders and the media’s purposeful often exceeds the number of voters in many countries, or accidental spread of false information. In Mexico, political debates dominated the online space and the the Council found automation and artificial amplifica- world of messaging platforms—disinformation and tion to be more prominent. Atlantic Council research- misinformation followed, intricately intertwined into ers uncovered actors who hired commercial bots for those conversations. financial gain and used political bots for the spread of A rapidly evolving information environment— specific electoral messages. At the state level, mean- wherein innovation often outpaces traditional security while, disinformation about the electoral process measures and governance at the private and public exacerbated polarization. levels—became a catalyst and a vector for the spread In response to the challenges posed by disinfor- of rumors and false information. Disinformation mation, stakeholders mobilized to strengthen digital actors, whether through organic disinformation or by resilience. Impressive, multi-stakeholder engagement employing artificial amplification, provoked fear and from electoral bodies, international institutions, tech- anxiety, and sought to illicitly influence voters, under- nology and social media companies, civil-society and mining the electoral process along the way. academic organizations, and the media comprised To address this complex set of challenges, the a series of nascent, growing, and evolving efforts. Atlantic Council’s Adrienne Arsht Latin America Lessons learned from 2018 will fuel solutions to the Center and its Digital Forensic Research Lab global challenge of disinformation for decades to (DFRLab) partnered with local organizations to iden- come. By highlighting some such lessons and laying tify, expose, and explain disinformation, to promote out potential solutions to be considered in the next increased dialogue, to support concerted action, and phase of this fight, the Atlantic Council today lays out to increase digital and media literacy as a bulwark of a vision for a future where facts and truth prevail. democracy. #ElectionWatch Latin America—as the Authentic communications underpin the legiti- effort was called—shed new light on the ways disin- macy and resilience of democracies around the world. formation, misinformation, and automation appeared Through a multi-stakeholder approach, society must within the context of each country’s election environ- continue to understand disinformation, to strengthen ments and influenced outcomes. digital resilience, and to transform the digital engagement space, not only ahead of future elections in Latin America, but ahead of elections in every corner of the globe.

Graham Brookie, Jason Marczak, Roberta Braga, Director and Director, Adrienne Associate Director Managing Editor, Arsht Latin America and Project Lead, Digital Forensic Center, Atlantic Adrienne Arsht Latin Research Lab Council America Center, (DFRLab), Atlantic Atlantic Council Council

3 disinformation in democracies

Carlos Reina (C), Introduction his mother Leida Castillo (R) and his sister Cyndia, huddle around Carlos’ mobile phone. Cell phones are increasingly the go-to medium for information consumption in Latin American countries.

he world is more connected than ever, with between a grandfather in Brazil being lauded or den- more than four billion people engaged igrated for making putty on YouTube. Unstopped, online, and 3.5 billion on social media. In false rumors can lead to the mob lynching of a young Latin America, where many countries are Indian man wrongfully accused of kidnapping children Tmobile first, online connectivity is widespread and in his village. escalating. In 2019, more than 101 million mobile- That moment when a person makes the deci- phone users in Brazil will access the Internet from their sion, from the comfort of his or her couch, to send a devices, up from around 72 million in 2015.1 seemingly real video about electronic voting fraud It is no surprise, therefore, that new challenges to followers and networks on WhatsApp, Twitter, or have emerged over the years in conjunction with the Facebook, may very well risk incremental and last- changing tides of information consumption and con- ing damage to the democratic ideals and institutions nectivity. Disinformation (false information spread underpinning freedoms and rights in their country. with an intent to deceive) and misinformation (false More than in any Latin American electoral cycle information spread without intent) have begun to to date, disinformation and misinformation made deeply influence communications. Today, critical their rounds on social media in Brazil, Mexico, and consumption of information, with an eye for disin- Colombia in 2018. False narratives, hyperpartisan blog formation and misinformation, can be the difference posts, edited images, and misleading videos fueled

4 REUTERS/Carlos Garcia Rawlins introduction

increased polarization as voters headed to the polls. Research Lab partnered to create #ElectionWatch Society raced to understand how and why disin- Latin America to identify, expose, and explain disinformation spread in the Western Hemisphere. Some formation where and when it happened around elec- blamed technology companies and social media tions in Brazil, Colombia, and Mexico. companies. Others blamed the political “establish- Going beyond fact-checking and working along- ment.” Yet others blamed the traditional news media. side partners in the region—including think tanks, Meanwhile, a spotlight turned on WhatsApp as the media, and universities—and in conversation with messaging platform dominated headlines in Brazil. technology firms and electoral authorities, the And, in the United States, Congress called hearings Atlantic Council’s #ElectionWatch Latin America to determine the feasibility and risk associated with researchers used open-source data (anything pub- potential Russian interference in Mexico’s elections. licly available online) to explain how cases of disin- While it may be tempting to blame a few organi- formation emerged, who was affected, who ampli- zations or platforms for the challenges associated fied stories, how they spread, and what came out of with disinformation and misinformation, instances of that circulation. such during the 2018 elections in Latin America were Rather than assuming its own credibility, through a byproduct of a confluence of factors. Economic #ElectionWatch Latin America the Atlantic Council downturns, unemployment, corruption, mistrust in harnessed open-source information to explain disin- politicians, and a desire for change came together formation with an element of transparency; the find- to create an environment ripe for the spread of ings and the tools used to conduct these investiga- falsehoods online. Illicit actors, networks, and social tions were available for anyone to verify. In parallel, media users bought and sold likes and shares on the council undertook capacity-building trainings social media, and journalists throughout the region that leveraged on-the-ground knowledge, and built struggled to get ahead of disinformation, while a robust and sustainable community of local nongov- simultaneously working to prevent the amplification ernmental actors in the information space. of false news. The findings of this year-long effort are high- In Brazil, as the country descended into an eco- lighted in this report. But the #ElectionWatch nomic recession and grappled with a wave of corrup- Latin America effort is part of a broad, concerted tion that implicated nearly two thirds of politicians effort to strengthen digital resilience in Latin in the lead up to the presidential elections, organic America and beyond. The challenges have not dis- disinformation spread like wildfire—at times, with sipated post-elections, and more must be done to politicians exacerbating disinformation and misinfor- strengthen a field ready to tackle changing commu- mation narratives. nication trends. The Atlantic Council, in conversa- In Mexico, Atlantic Council researchers and part- tions with partners on the ground, takes a first step ners detected automation in national and local pol- to outline suggested recommendations to protect itics. Moguls like Carlos Merlo thrived on a disecon- public debates from those who would seek to under- omy of disinformation, monetizing the buying and mine them in the digital-engagement space. selling of likes and shares on Facebook and Twitter. Digital resilience is not an issue to be approached And in Colombia, only a short time after the refer- by one or two organizations. Truly global challenges endum on the proposed peace agreement between by their very nature, disinformation and misinfor- the government and the Revolutionary Armed mation must be addressed through the collective Forces of Colombia (FARC), a polarized race gave efforts of governments, technology companies, tele- way to blame games between the right and the communications organizations, civil society, media, left, with each side using disinformation to attack and academia. the other. Media, in their attempt to combat it, often As the world dives ever more deeply into the sea unintentionally amplified disinformation. of online engagement, now is the time to ensure that At the core of the Atlantic Council’s mission is democratic ideals and freedoms are upheld, while the belief that democracy relies on debate. In any taking action to prevent illicit online activity. Any democracy, facts are fundamental to that produc- sustainable solution to the problem of disinforma- tive debate. Unfortunately, the massive consumption tion and misinformation will need to come from hon- of unbalanced or intentionally misleading informa- est conversations among diverse actors, including tion online has the potential to throw democracy off in the private and public sectors, media, and civil balance. society. This report seeks to support these actors, to This is why, in 2018, the Atlantic Council’s Adrienne strengthen a truly multi-stakeholder approach. The Arsht Latin America Center and its Digital Forensic time for digital resilience is now.

5 disinformation in democracies

DISINFORMATION IN THE 2018 ELECTIONS

BRAZIL By Luiza Bandeira and Roberta Braga

key findings Disinformation in Brazil’s Hyperpolarization Countering disinformation elections included organic and a lack of trust in narratives, rather than disinformation spread institutions—spanning instances of disinformation, through encrypted government, media, may be more impactful messaging platforms. and civil-society in assuring that organizations—created disinformation does not set an atmosphere in the agenda for electoral which disinformation debates and media spread quickly, with coverage. compounding effects.

CONTEXT

On October 28, 2018, Brazilians elected former Army Captain Jair Bolsonaro as their next president, following a polarizing and contentious race. Far from ordinary, the 2018 general election took place at a unique point in Brazilian history, one that followed a deep economic recession and a political crisis that saw scores of politicians jailed or investigated for corruption. The economic and political crises accentuated violence and insecurity—in 2018, petty crime soared, and homicide rates hit a record high of thirty per one hundred thousand people.2 Taken together, these factors fostered widespread rejection of the established political parties—especially the Workers’ Party (Partido dos Trabalhadores, or PT), which had governed Brazil for most of the previous fifteen years. They also helped elevate change candidate Jair Bolsonaro, a legislator previously on the sidelines of Brazil’s political scene, known mainly for his controversial statements during nearly thirty years in Congress. Bolsonaro was elected with 55 percent of the vote in the runoff election—a victory attributed not only to the complicated economic and political moment in which Brazil found itself, but also to his effective use of social media as a primary means of connecting with Brazilian voters.3

6 brazil

Citizens mobilize in support of the Lava Jato (Car Wash) investigations in São Paulo, Brazil. Corruption scandals contributed to record-low levels of trust in Brazil in 2018.

Economic Stagnation and Brazil’s and his subsequent barring from the 2018 election),4 Operation Car Wash and the party’s perceived inefficiency in addressing Brazil’s crises over its years in power, led to notable o understand how Brazil’s economic stagna- rejection of the “left” and the PT in the 2018 presiden- tion (marked by a shrinking of the economy and tial election. Tpeak unemployment), political mistrust (fos- Jair Bolsonaro faced off in a contentious race tered by the Lava Jato or Operation Car Wash cor- against Fernando Haddad, who took over as the PT ruption investigation), and insecurity (exacerbated candidate after Lula was barred from running. Though by rising crime rates) intertwined with deep distrust Lula led in polls before being removed from the race, in a hyperconnected, mobile-first society to shape Bolsonaro took the lead against Haddad. Bolsonaro debates and the spread of disinformation around presented himself as an outsider and ran on a platform Brazil’s election, it is important to consider sentiments that proposed liberal economic policies and conserva- in Brazilian society leading up to the 2018 elections. tive values as a way of undoing the progressive poli- Between 2013 and 2018, debates in Latin America’s cies implemented by the Lula and Rousseff adminis- largest democracy became increasingly polarized. trations. He also advocated a strong security platform Mass street protests in 2013 sparked ever-widen- that included easier access to guns for Brazilians. ing divisions during and after the 2014 presidential race that saw President Dilma Rousseff narrowly Lack of Institutional Trust reelected. As Brazil became engulfed in political and economic crises over the next five years, divides spite having little financing or television and became starker. The crises, mutually reinforced by radio air time—two components that had and intertwined with Operation Car Wash, signifi- Dlargely determined presidential victors in the cantly affected the PT. The impeachment of President past—Bolsonaro effectively used social media as his Dilma Rousseff of the PT party during her second primary means of connecting with Brazilian voters term in 2016, the arrest of former President Luiz already deeply distrustful of politics as usual.5 In 2018, Inácio Lula da Silva (the most visible face of the party Brazilians’ trust in institutions reached its lowest level

REUTERS/Amanda Perobelli 7 disinformation in democracies

in ten years. Among the least trusted institutions Almost 90 percent of Brazilians watch television to were the presidency and political parties, while the stay informed, and television is the most important police, churches, and the Armed Forces were among outlet for 63 percent of Brazilians—four broadcast- the most trusted.6 ers (Globo, SBT, Record, and Band), dominate the Bolsonaro’s messages, which offered solutions to Brazilian broadcast market, capturing more than 70 the many issues Brazil faced and resonated with wide percent of the Brazilian audience per 2017 numbers.12 swaths of the population, reached voters on social While Internet use is on the rise and television con- media and messaging platforms, where they increas- sumption rates remain high, print newspapers are los- ingly absorb information. ing ground.13 Only 3 percent of Brazilians use newspa- The degree of distrust, combined with the high lev- pers as their main medium for staying informed, and els of digital engagement in the country’s information 12 percent say they use newspapers along with other ecosystem, created fertile ground for the spread of media. Between 2014 and 2017, the circulation of the disinformation and misinformation around the elec- eleven most widely sold print newspapers fell by 41 tions. In 2018, disinformation and misinformation percent, to 736,346—the number of digital subscrip- spread in an organic, domestic, hyperpartisan way.7 tions has increased marginally, but does not com- pensate for the loses in print circulation.14 As a result, since 2017, Brazilian newspapers have tightened their paywall policies. In Brazil’s information space—as, increasingly, in INFORMATION AND other countries throughout the world—it is important to note there is now a gap between where news is ELECTION ECOSYSTEMS now published and where information is consumed. With newspaper consumption in decline, readers To understand how disinformation and misinforma- increasingly use messaging platforms like WhatsApp tion permeated communication channels during to access information. Despite this trend, professional Brazil’s 2018 elections, it is important to understand journalism still does not extensively publish news on how Brazilians consume information, as well as how the platform, as they have not yet found a sustainable the country’s electoral system functions. business model for use of the application.

Brazil’s Information Ecosystem he Brazilian media market is highly concen- trated, both in terms of audience and owner- TABLE 1 Tship. Although the market is dominated by large commercial television broadcasters, the coun- Media Consumption in Brazil try is also one of the most connected on social media What is your preferred medium for and mobile applications.8 getting informed about what happens in One hundred and twenty million people have Q Brazil? What is your second option? access to the Internet in Brazil, using it mainly to exchange messages and access social media. In 2017, for the first time in history, the number of Brazilians Medium 1st option 2nd option accessing the Internet only via mobile outperformed the combined use of computers and cell phones. Television 63 89 In 2017, a study showed that Brazil had 310 million Internet 26 49 active smartphones, more than its population of 210 million people.9 Notably, mobile-only access was Radio 7 30 more prevalent among the country’s poorer popu- Newspapers 3 12 10 lation. The Internet was mentioned by 49 percent Magazines 0 1 of people as a means of staying informed, and by 26 percent as their main source of information. Social Outdoor advertisements 0 0 media played, and continues to play, an important (billboard, bus, elevator, role in the flow of information in Brazil. According to a metro, airport) Datafolha Institute survey that focused exclusively on Other 0 2 social media, WhatsApp was used by 66 percent of Doesn't know/didn't reply 0 0 Brazilian voters, followed by Facebook (58 percent), 11 Instagram (36 percent), and Twitter (14 percent). SOURCE: "Pesquisa Brasileira de Mídia 2016—Hábitos de Consumo de Mídia pela População Brasileira", pela Though online engagement is prevalent, television Secretaria de Comunicação Social do governo remains the most consumed medium in the country.

8 brazil

Brazil’s Election Ecosystem DISINFORMATION AND oting in Brazil is mandatory for literate citizens between the ages of 18 and 70; in 2018, 147 mil- MISINFORMATION IN lion people were registered to vote in the coun- V15 THE 2018 ELECTIONS try. Presidential elections take place every four years. Presidents are elected by the majority of voters in a two-round system, with the possibility for reelection To inform, expose, and explain disinformation and to a second term. Along with presidential elections, misinformation ahead of and during the electoral the country votes for members of Congress and state cycle in Brazil, the Atlantic Council partnered with governors every four years. leading think tank and higher education institution Brazil is a multiparty system; in the 2018 general Fundação Getulio Vargas’ Department of Public Policy election, the country had thirty-five registered par- Analyses (FGV DAPP)—whose Digital Democracy ties.16 Parties receive public funding for elections and, Room23 reinforced strengthening of digital resilience since 2015, corporate donations have been banned.17 in Brazil—and leading news outlet Jornal O Globo. As Notably, Brazil has relied on an electronic voting sys- part of #ElectionWatch Latin America in Brazil, the tem for more than twenty years, since 1996.18 Atlantic Council conducted research over the span Elections at the municipal, state, and federal levels of three months—from August to October 2018—and in Brazil are organized and monitored by the Justiça published on an ongoing basis. Atlantic Council find- Eleitoral (Electoral Justice). The most important elec- ings showed fringe partisan websites emerging and toral authority is the Tribunal Superior Eleitoral (the contributing to hyperpartisan debates, organic dis- Superior Electoral Court or TSE), and each state has information spreading at high rates, and the role of its own state court.19 The Electoral Justice is respon- WhatsApp as a vector for disinformation. sible for including voters in the voter’s registry, monitoring parties’ fundraising and expenditures, and The Increased Relevance of Partisan verifying who can run for office. The TSE determines how much free campaign time on television and radio Media Versus Traditional Media each party will have, based on its representation in the s polarization grew in Brazil, so too did hyper- Chamber of Deputies, the lower house of Congress.20 partisan websites. In the past, these web- The Electoral Justice handles the logistics of elections, sites have been called “fake news” web- A24 including choosing polling places, training personnel sites. However, this designation does not accurately and distributing electronic voting machines through- describe these websites, which combined “decon- out the country. The Electoral Justice also deals with textualized truths, repeated falsehoods, and leaps of electoral lawsuits, such as electoral crime and inspec- logic to create a fundamentally misleading view of the tion of campaign advertising. world,” rather than publishing completely fabricated The activities of the Electoral Justice are moni- stories—as was the case in the US election of 2016. 25 tored by the Ministério Público Eleitoral (the Electoral By adopting sensationalist headlines and appealing Public Ministry, also known as the Electoral Public to emotions, these websites became increasingly pop- Prosecutor’s Office), which can initiate legal action ular. Atlantic Council investigations conducted as part in cases concerning parties and candidates, in addi- of #ElectionWatch research in Brazil found, for exam- tion to monitoring the legality of voting at polling ple, that partisan websites outperformed traditional stations.21 The Electoral Public Ministry oversees the independent media in the corruption debate in the six legitimacy of various components of the electoral pro- months that preceded the electoral campaign.26 cess, including voter registration, party conventions, At least one hyperpartisan website benefited from registration of candidacies, campaigns, and electoral inauthentic amplification to spread its messages. The propaganda. Both federal and state law enforcement website República de Curitiba (Republic of Curitiba, investigate, and act to prevent, electoral crimes.22 a reference to the city where the Lava Jato corruption investigation had its headquarters), a pro-Bol- sonaro blog, was the site with the second highest rate of engagements during the elections. In December, Facebook took down the República de Curitiba page due to inauthentic coordinated behavior—online behavior intended to deceive audiences about its authors or motives, conducted by several users in coordination with one another (see table 2). This inauthentic behavior is banned on Facebook. While hyperpartisan websites flourished, independent media outlets were accused of publishing “fake news” during the elections. This criticism was led by then-candidate Bolsonaro himself, a vocal critic of

9 disinformation in democracies

mainstream media who has referred to some news outlets as “fake news.”27 One week before the first round of voting, protests against Bolsonaro took place. Posts by Bolsonaro TABLE 2 supporters and pro-Bolsonaro pages on Twitter and Facebook claimed the media were artificially inflating Top Domains on All Networks the number of protesters at the rallies. Some users and Image shows the media outlets users engaged most pages shared edited images and pushed narratives with during the election in Brazil. campaign. suggesting the same images were being shared by the

news media. These users and pages also disseminated Total the idea that the Jornal O Globo newspaper had pub- Domain Engagements lished a picture of a different manifestation claiming it was from the anti-Bolsonaro rally. However, the pic- noticias.uol.com.br 85,714,716 ture was authentic, as Atlantic Council #ElectionWatch republicadecuritiba.net 33,955,680 research showed.28 exame.abril.com.br 30,467,322 Independent media were also accused of manipulation in one of the most significant cases of elec- oglobo.globo.com 28,663,215 toral violence registered during the election period. veja.abril.com.br 27,766,855 Capoeira Master Moa do Katende was murdered on www.youtube.com 21,849,733 October 7, 2018, on the night of the first round of elections. The suspect, a Bolsonaro supporter, was arrested www.revistaforum.com.br 19,924,580 and, according to law enforcement, confessed his politica.estadao.com.br 18,916,390 political motivation for the crime—the victim sup- www.terra.com.br 17,590,272 ported PT candidate Fernando Haddad. However, while in front of television cameras, the suspect denied www.oantagonista.com 17,551,897 this political motivation. Per Atlantic Council findings, www.infomoney.com.br 15,727,150 some independent news outlets failed to effectively relay his version of the story and were accused of pub- noticias.r7.com 15,501,399 lishing “fake news” by partisan websites that claimed istoe.com.br 15,304,067 the murder was not politically motivated. These parti- brasil.elpais.com 14,191,831 san websites, however, ignored the police’s version of events, failing to tell the entire story.29 g1.globo.com 13,680,766 www1.folha.uol.com.br 13,608,318 Coordinated Organic Digital br.noticias.yahoo.com 10,911,533 Engagement Prompted by Political cbn.globoradio.globo.com 9,411,535 Divisions www.esquerdadiario.com.br 8,991,078 s disinformation and misinformation emerged, researchers found it difficult to establish origi- catracalivre.com.br 8,370,107 nal authorship and to quantify how organized www.cartacapital.com.br 7,572,987 A the spread was during the elections. The Atlantic www.diariodocentrodomundo.com.br 6,041,835 Council used open-source methodologies to establish who were the main amplifiers of these messages jovempan.uol.com.br 4,992,469 on open social media platforms such as Facebook and www.brasil247.com 4,319,342 Twitter. Atlantic Council research found that the elec- noticias.band.uol.com.br 3,892,767 toral race in Brazil was marked primarily by organic online engagement. Disinformation and misinforma- www.noticiasmacae.com 3,543,493 tion generally garnered traction after being shared by www.gazetadopovo.com.br 3,539,154 “influencers” whose messages were then further disseminated by their followers. lula.com.br 3,313,480 Flávio Bolsonaro, Jair Bolsonaro’s son, and Joice www.valor.com.br 3,282,784 Hasselmann, a former journalist and congresswoman, congressoemfoco.uol.com.br 3,247,737 were two influencers that took part in the spread of videos wrongly suggesting fraud in electronic voting SOURCE: Buzzsumo machines, per a Tribunal Regional Eleitoral of the state of Minas Gerais investigation in the case of the former and a Tribunal Superior Eleitoral conclusion in the case of the latter. And, after Jair Bolsonaro was stabbed, both his followers and his opponents spread false

10 brazil

Supporters of Brazil’s then- President-elect Jair Bolsonaro travel to Brasilia to watch the president’s inauguration ceremony. Jair Bolsonaro, who portrayed himself as the change candidate in Brazil’s 2018 elections, took the helm after a polarized election.

WhatsApp as a Vector for Disinformation information on Facebook groups and pages—while he messaging application (app) WhatsApp was the left claimed Bolsonaro had staged that attack, a vector for disinformation in the 2018 Brazilian the right claimed the attacker was a member of the elections. While discussions about disinforma- 30,31 T Workers’ Party. Both claims were false. tion in Brazil focused on WhatsApp and its poten- These messages were then shared by the influenc- tial influence on voters’ choices, because the app is ers’ followers. Bolsonaro’s supporters, in particular, encrypted and has no public API, there are no com- were very active on social media.32 They often acted prehensive metrics available to verify the extent to in coordination, though not necessarily using automa- which disinformation or misinformation was shared tion or artificial amplification by bots or botnets. This within the platform. Per WhatsApp, the platform also became evident, for instance, during the migration of does not have a search function and is not built for Bolsonaro supporters to the alt-right social media plat- users to garner large followings. Per 2017 estimates, form Gab.ai. The most retweeted post read “Let’s up the application is used by at least 120 million people the hashtag #MeSeguenoGab (follow me on Gab).” in Brazil—according to 2019 data from the company, The hashtag had twenty-five thousand mentions, and WhatsApp has more than 1.5 billion users worldwide. no indication of automated amplification was found.33 The app utilizes end-to-end encryption, enabling #ElectionWatch Latin America research also detected only the sender and receiver of a message to access campaigns on Facebook, Twitter, and WhatsApp its content—the mechanism is aimed at preserving prompting users to boycott the media and to “dislike” privacy and security.35 The popularity of the app in videos on YouTube.34 Brazil and in other countries, nonetheless, is partly Bolsonaro himself was very active on social media. attributed to economic reasons, rather than only to On Facebook, for instance, none of the 12 other can- privacy concerns—telecommunications companies, in didates who ran for the presidency came close to partnership with Facebook, adopt “zero-rating” pol- his audience engagement and reach. Bolsonaro fin- icies that allow users to access the platform without ished the race with 8.6 million “likes” on his official tapping into their mobile data plans.36,37 This is also Facebook page—runner-up Fernando Haddad gar- sometimes the case for other social media companies. nered 1.8 million (see table 3). Being a messaging platform rather than a broadcast

REUTERS/Amanda Perobelli 11 disinformation in democracies

TABLE 3 "Likes" on Brazilian Presidential Candidates' Facebook Pages (in Millions)

Graph shows comparison among candidates. 7.95 2.39 1.68 1.11 0.77

Jair Bolsonaro Marina Silva Fernando Haddad Geraldo Alckmin Ciro Gomes

SOURCE: Buzzsumo

social media platform, WhatsApp has no timeline or to find a sustainable business model for news on newsfeed comparable to those on Facebook and WhatsApp.39 This means journalists end up using tra- Twitter. Each piece of content a user receives on ditional channels (their websites and social media) to WhatsApp is sent directly to them or to a group of publish news, while readers flock to a different chan- which they are a part. This means the app is perceived nel (WhatsApp) to get informed, resulting in a gap as a private environment to communicate with friends between producers of trustworthy news and readers. and family. This gap can be exploited. Nevertheless, the app also has broadcast features Second, because access to websites outside that allow for broader communication and that at zero-rating data costs money, there is a disincentive to times enable messages to go viral. By the first round share more informative links (like news articles) and an of elections on October 7, 2018, each user could cre- incentive to share less informative content (like images ate up to 9,999 groups,38 each with up to 256 peo- and memes). Additionally, many people avoid click- ple, or single lists that broadcasted a message to 256 ing on links that would take them out of WhatsApp to contacts at the same time. Users could also forward verify messages, especially since these messages are a message to up to twenty contacts simultaneously. often sent by friends and family who are considered Shortly before the campaign kicked off in August 2018, trustworthy sources. This reduces the likelihood that WhatsApp added a label that identified forwarded readers will use external sources to verify information messages; prior to that, users had no way of knowing they receive on WhatsApp.40 whether the author of a message was the person who Finally, it is difficult to track the spread of a message sent it or if this person had forwarded the text from on WhatsApp at the scale or speed at which content one of their contacts. In December 2018, WhatsApp circulates. Currently, because of the way WhatsApp is restricted forwarding from twenty to five contacts. designed, the messaging platform does not allow for The widespread use of WhatsApp and its perceived automatic notifications to readers alerting them that privacy—because groups and forwarding features the information they are sharing is false, for tracking sometimes result in users not realizing a message sent the performance of articles debunking false news, for could be circulated widely within other networks on the mapping of how narratives are amplified, or for WhatsApp—combined with encryption and zero-rat- complete removal of content with false information ing policies, represents a triple challenge in the infor- from the app. Importantly, because of encryption, it is mation environment. difficult for researchers to use WhatsApp data to iden- First, in Brazil’s information space, as noted in the tify who the first person was to post a message con- previous section, there is a gap between where news taining false or misleading information.41 is published and where information is consumed. Thus, researchers and journalists in Brazil focused Readers increasingly use WhatsApp to access infor- on WhatsApp messages shared by consum- mation, but professional news outlets do not pub- ers with fact-checkers and on “public” or “open” lish news on the platform in any consistent way. Since groups—groups of up to 256 people that could be there is no effective means of tracking how a piece of accessed through links shared on the Internet, usu- information is shared nor to obtain metrics to compre- ally on Facebook.42 Of the public groups analyzed hensively verify audience engagement, it is difficult by researchers, many were connected to Bolsonaro

12 Official facebook pages of Brazilian Presidential Candidates brazil

WhatsApp public groups often supporters, with his own family being part of some. function as a messaging pyramid While WhatsApp did not disclose the exact number of public groups that exist, the company indicated scheme. Group members receive that worldwide approximately 10 percent of messages material that they disseminate are sent in groups comprising up to 256 people.43 According to the company, most groups have fewer to their personal networks, than ten people. During the election cycle, WhatsApp who then share it with other public groups often functioned like messaging pyramid schemes. Group members received material networks, rippling distribution that they disseminate to their personal networks, who and amplification to a much shared it with other networks, rippling distribution and amplification to a much larger audience.44 larger audience. While it was not possible to access complete data about the spread of disinformation within the encrypted platform, by tracing the most-shared mes- Narratives SOURCE: Buzzsumo sages on public groups, researchers and fact-checkers were able to show that those widely shared were lectionWatch research in Brazil showed the either false or taken out of context. Moreover, a study main narratives circulating on WhatsApp were by Brazil’s Instituto de Tecnologia e Sociedade, or ITS Emost often not different from those dissemi- Rio, found users spreading a high number of mes- nated on Facebook or Twitter. This means WhatsApp sages in short intervals, as well as profiles with no is not necessarily the beginning or the end of false names and impersonal profile pictures, suggesting information. Rather, it works as a part of the informa- automation was used in these public groups.45 tion ecosystem, in which disinformation spreads.47 It In addition to the public groups, media reported also implies that debunking narratives, rather than spam being shared with thousands of WhatsApp debunking individual messages, might be more fruit- users. Though it is still unclear whether these mes- ful to preventing the spread of disinformation and sages contained disinformation, this has prompted misinformation. discussions about the possibility of the app being used to circumvent campaign finance laws.46

Narrative Background Examples of False Information Electoral fraud Independent analyses of Brazil’s electronic Videos on WhatsApp showed voting machines voting system found vulnerabilities that allegedly forcing voters to cast a ballot make the system susceptible to a potential for Fernando Haddad, the Worker’s Party hack. In twenty years, no consistent candidate.49 evidence of fraud has emerged.48 Child In 2011, a government program on public Candidate Fernando Haddad, who was the sexualization and health aimed at promoting acceptance and minister of education when the project was “gender ideology” preventing homophobia in schools was discussed, was accused on WhatsApp of being cancelled after conservatives denounced a pedophile and a supporter of incest. Messages it as a “gay kit” intended to sexualize also claimed that, if he were elected, children children.50 would become the property of the state at the age of five and their gender would be chosen then.51 Anti-left and anti- The left-wing PT governed Brazil for thirteen Images were circulated of a check that was Workers’ Party years and saw many of its politicians supposedly part of a PT corruption scheme. messages accused of corruption. The country was also Messages alleging the PT was communist and engulfed by an economic crisis toward the Brazil would become the next Venezuela were end of the PT’s mandate.52 also shared widely.53 Anti-media Traditional media faces a crisis of trust. Messages claimed articles published by the media content Bolsonaro continues to be a vocal critic of were wrong. Campaigns called for the boycott of the media and often claims they publish media outlets.55 “fake news.”54

13 disinformation in democracies

TABLE 4 Word cloud shows words like “Bolsonaro,” “Haddad” and “TSE” related to the expression “fake news.”

SOURCE: Sysomos.

Although it is not possible to determine who spread Although it is not possible to affirm with certainty these messages at scale—because of the encrypted how disinformation and misinformation affected votes, nature of WhatsApp and due to the pyramid-scheme both did cause unrest at the national level during the design of groups—then-candidate Jair Bolsonaro was election cycle. The most visible example was the spread one, if not the main, beneficiary of these narratives. of the electoral fraud narrative, which was worrisome Research has shown that the “gay kit” and the “elec- because it further eroded trust in institutions. Per toral-fraud” stories, which clearly targeted Haddad Atlantic Council #ElectionWatch Latin America report- and bolstered Bolsonaro’s campaign messages, were ing, suspicions about electoral fraud gained momen- the most shared messages on Twitter, Facebook, and tum when Bolsonaro published a video on his Facebook YouTube.56 However, the then-candidate was also the page claiming the possibility of fraud was concrete.59 victim of rumors, most connected to his stabbing, which This happened around the same time polls indicated left-wing groups claimed had been staged to garner Bolsonaro was losing support ahead of the first round popular support. of voting on October 7. The day of the first round of voting was marked by claims of electoral fraud, many false. One of the more widely shared videos showed a voting machine allegedly forcing voters to cast a ballot for Haddad, the IMPACT ON THE Worker’s Party candidate. This video was shared by one of Jair Bolsonaro’s sons on Twitter and quickly gained ELECTIONS traction, despite statements made by Brazil’s electoral court affirming the video had been manipulated. he direct influence of disinformation on voter At polling stations across Brazil, voters claimed voting preferences in the Brazilian elections is diffi- machines were not working and registered formal com- cult to quantify. Two opinion polls conducted in plaints with the Electoral Justice. Some claims were T 60 October showed different results. After the first round, dismissed by the police. The Federal Police and the an Ibope poll suggested three in four voters did not Electoral Justice, however, did not publish a full report receive negative messages about any candidates on with the outcomes of all investigations, making it diffi- WhatsApp the week prior to voting. Hence, per Ibope, cult to examine the scale of the problem and the extent the impact of WhatsApp was limited.57 Another poll, by to which the problems involved any kind of fraud. Idea Institute, suggested that 98 percent of Bolsonaro Similar claims of electoral fraud circulated around the voters were exposed to false messages and 90 percent second round of voting. One person was arrested in the believed them.58 Opinion polls also might be affected state of Rondonia after publishing a video telling peo- by misreporting. ple to use glue to break voting machines to stop people

14 brazil

ITS Rio: Net Neutrality and Data Protection in the Fight Against Disinformation

Ronaldo Lemos, Co-founder and Director, Institute for Technology & from voting for the PT.61 The president of Brazil’s top Society (ITS Rio) electoral court also received online threats.62 Disinformation also impacted Brazil’s election by disinformation online is a global concern and setting the agenda for media coverage and political one that stems from various factors. In Brazil, debates online. Candidates and news outlets spent strengthening personal data protection is a countless hours responding to rumors and false news way to prevent the misuse of personal data and, in the process, spent less time covering the plat- for political purposes, and to empower users forms being proposed by the candidates, and their his- to exercise more control in how their data tory and achievements as politicians.63 For example, is used for economic purposes. In Brazil, a during the campaign period (August 15 to October 28), General Data Protection Law was approved in the term “fake news” was mentioned 1.9 million times August 2018. However, implementation of the on Twitter. The numbers illustrate how disinformation law remains a challenge. As of March 2019, no and false narratives became something to which candi- final definition of the structure of Brazil’s Data dates, media, and the voters had to react. Protection Authority existed. One concern Finally, disinformation and misinformation likely is whether that authority will have enough increased distrust in traditional media (see table 4). The independence from the government. Having perception that the established media fabricated news an independent authority is an important to harm Bolsonaro’s credibility is a worrisome trend element to establishing the compatibility of that did not cease after the campaign. This makes it Brazilian law with the European General Data harder for journalists to counter partisan information Protection Regulation. and government propaganda. Brazil took a long time to enact data protection legislation. As a result, enforcement of data protection in the country has been sporadic, and there is an overall absence of a data protection culture. As INSTITUTIONAL such, government databases have been made available voluntarily or leaked to private REACTIONS entities without deep scrutiny. Brazilians often find their personal data, including tax enrollment numbers, identity numbers, Government personal address, phone number and other razilian authorities began mobilizing to address sensitive information, freely available online the challenges posed by disinformation as in privately-owned but publicly accessible Bearly as 2016. In 2017, Brazil’s Electoral Justice databases such as “Tudo Sobre Todos.” approved a resolution on election-related advertis- Naturally, this data ends up in the hands ing that laid out specific rules for political advertising of political campaigns and disinformation on the Internet and on social media. 64 The resolution actors, and at times is used by illicit actors for defined, for instance, that two types of advertisements intimidation and coercion online. As of March on the Internet were permitted: the boosting of and 2019, Brazil’s data protection law comprised sponsoring of links, such that a candidate or party’s a loophole that enabled government bodies content would become visible to a greater number of to freely transfer data to private entities, by users. However, campaigns could only boost or spon- means of a simple “partnership” (“convênio”) sor links under two conditions: first, the content had without further scrutiny or guidelines. to be identified as electoral content associated with a ITS Rio believes good data protection laws candidate or political party, for transparency reasons; should help curb the abuse of personal data second, the hiring of these types of amplifications for the purpose of disinformation and other online could only be done by a candidate or political weaponized forms of communication in Brazil. party, or by their official representatives.65 Before the elections, Brazil’s top electoral court

15 disinformation in democracies

(TSE), in committing itself to standing strong against apparatus to benefit its candidates. There is, how- disinformation, created a group that gathered law ever, no consensus about what information can and enforcement and Internet specialists to discuss the cannot be published. In the case of Moa do Katende’s issue. On August 22, then-TSE President Luiz Fux murder, the secretary of security of Bahia failing to announced candidacies involved with “fake news” publish any official information about the case online could be barred from running.66 probably contributed to the spread of inaccurate During the campaign, however, the court failed to information as this action prevented citizens who adopt strong measures against disinformation. The wanted to independently verify information from group met only six times, and not at all in the three accessing statements from the official source.72 months preceding the election. The new president of TSE, Rosa Weber, said she did not know what the Technology and Social Media former president had said about the issue and stated there was no “easy solution for the problem.”67 Companies On the day of the first round of voting, fact-check- head of the Brazilian elections, technol- ers criticized the TSE for failing to provide answers ogy and social media companies took some to their questions, which left rumors unanswered. Aaction to address the challenges posed by Prior to the second round of voting, the TSE met disinformation. Facebook partnered with local with fact-checkers, and technology and social media fact-checking agencies, such as Aos Fatos and Lupa, companies, to discuss disinformation, resulting in to identify false information on the platform—the the creation of a task force the weekend of the sec- company worked to reduce the reach of posts classi- ond round, during which time fact-checkers had con- fied as false by fact-checkers.73 Facebook also began tact with the legal and technology arms of the TSE.68 labelling political advertisements and launched its The court also launched a website to address rumors archiving capability. Brazil became the second coun- related to the electoral process.69 Notably, the TSE try after the United States where Facebook imple- also responded to disinformation by ordering the mented these capabilities.74 removal of false accusations from the Internet, after Messaging platform WhatsApp, in June 2018, being summoned by parties.70 began labeling “forwarded” messages to clarify the At the state level, one of the electoral courts pub- original author of the message was not necessarily lished one of the most effective debunking efforts the person who shared it.75 WhatsApp also reduced of the election. An expert from the court analyzed a the number of users with whom a person could share video depicting alleged fraud at a voting machine to a message—from two hundred to twenty contacts prove it had been manipulated. The court was proac- (it would, following the election, further reduce that tive in debunking the video, which had been shared number from twenty to five). The messaging plat- by important politicians, including one of Bolsonaro’s form also partnered with fact-checking initiatives, sons. 71 including Comprova and ÉNóis. Along with these Finally, the Atlantic Council also observed that, initiatives, WhatsApp internal investigations also due to alleged electoral law restrictions, state sec- resulted in the suspension of “hundreds of thou- retaries—the equivalent of state-level ministers— sands” of accounts.76 It also reacted by implementing were not publishing information on their websites. a campaign disseminated through traditional media Brazilian law prevents governments from publish- and social media advising people about disinforma- ing some information during the electoral period to tion and advocating against the spread of disinfor- prevent parties in power from using the government mation and misinformation.

Ahead of the Brazilian elections, technology and social media companies took some action to address the challenges posed by disinformation.

16 brazil

FGV DAPP: Brazil’s Election was a

Turning Point for Social Media certainly be a challenge in the coming years. By Marco Aurelio Ruediger, Director, Department Monitoring done en masse of Public Policy Analysis, Fundação Getulio during the electoral period Vargas (FGV-DAPP) indicates that social networks have become influencers of the political debate. Although this brazilians approached the 2018 the electoral process in both level of influence enables the electoral cycle sure of the impact national and international press. amplification of arguments and of social media on politics, but The association of the macro actors, so effective action is still unsure about the level of themes monitored with the necessary to ensure democratic effect it would have on shaping presidential candidates revealed institutions can continue to world views in both the short that corruption was the most function properly and effectively. and long term. The key issue frequent subject discussed in DAPP points to the following presented to those interested debates about the election on as actions that can be taken to in deciphering the network Twitter. Following corruption, strengthen digital resilience and society was whether almost the network’s users were also preserve democratic legitimacy: real-time monitoring and more concerned with public security -Social media platforms should qualified analysis of social media and the economy. accompany the observation would positively influence the In this vein, partnerships of legal frameworks through political environment and favor with organizations such as the electoral campaigns on the a fairer (more informed and less Atlantic Council were essential Internet and provide research polarized) election. This effort to bring complementary centers with more access to data led to the creation of the Digital methodological approaches to and efficient communication; Democracy Room, #observa2018, the forefront, and to address -The Electoral Court System by the Department of Public challenges along the way. should promote mandatory Policy Analysis at Fundação The presence of automated and continuous communication Getulio Vargas (FGV-DAPP), accounts was recurrent in this with social media platforms, a pioneering initiative to period—influencing up to 12.9 empowering civil society monitor the public debate on percent of the interactions organizations to monitor and the web and identify acts of on Twitter in the final week analyze; disinformation, threats, and of the first round, according -The state and the civil society illegitimate interference in the to FGV DAPP’s methodology. must obtain more knowledge, political process. And these contributed to the through the promotion of Over the course of one spread of disinformation and research, about the connection hundred days, DAPP analyzed misinformation both inside and between public opinion and the more than 130 million tweets and outside of the country. Inflamed social networks used during the 163 million interactions on the by the scenario of political 2018 Brazilian elections. Facebook pages of the thirteen polarization, disinformation The 2018 Brazilian election was presidential candidates and other was used as a campaign tool undoubtedly a major turning key political actors, in addition by all sides involved in the point for the impact of social to public data from Instagram elections, reaching boundaries media on political processes. and YouTube—the latter, a still unknown and, therefore, not Therefore, democratic societies platform that proved to be fully understood. Such was the must work to better understand increasingly important during the case of the WhatsApp platform, the phenomenon, to avoid electoral race. The monitoring which was used as a vector for the corrosion of traditional resulted in 123 analyses, with the spread of false information institutions by disinformation and fourteen weekly reports, which inside an almost untraceable the misuse of technology. contributed to discussions about environment—something that will

17 disinformation in democracies

Fact-checking Agencies and the Media Fake), that comprised all outlets belonging to the group. A joint effort, established during the weekend t times in partnership with technology and of the second round of voting, also brought together social media companies, fact-checking orga- the country’s main fact-checking agencies: Aos Fatos, Anizations and the media also mobilized to Lupa, Boatos.org, E-farsas, Comprova, and Fato ou address disinformation and misinformation around Fake. Together, they debunked fifty rumors in for- the elections. A group of twenty-four media outlets ty-eight hours.77 In parallel, fact-checking agency came together to form the fact-checking initiative Aos Fatos introduced “Fatima bot,” a Twitter bot, Comprova, spearheaded by the journalism non-profit also available on Facebook Messenger, that replied First Draft, which functioned in a decentralized man- to and alerted users when posts they shared online ner, with fact-checkers working from their newspa- were inaccurate.78 per’s offices. WhatsApp granted Comprova access At least three initiatives in Brazil took fact-checking to the messaging platform’s application program- to WhatsApp: Aos Fatos, ChecaZap and O Poder de ming interface (API), enabling the group to receive Eleger. The three organizations crowdsourced pos- from the public questions about messages they sus- sibly false information from users, who submitted pected to be false on an intermediary website—with these by send them to the organization’s WhatsApp access to the API, the messages Comprova received numbers. After fact-checking some, the organiza- went to a server, and not to a phone, facilitating the tions sent the results of the verification back to users debunking process. Comprova published their veri- via broadcast lists of up to 256 subscribers each on fications on their website and on social media, and WhatsApp. replied to users who sent false messages to the Despite their efforts, fact-checkers and the media fact-checking service with the verification of the encountered various challenges in their efforts to message the user sent. address disinformation. For one, these stakeholders Globo, Brazil’s leading media group, also carried were not always able to disseminate their verifica- out a fact-checking effort, “Fato ou Fake” (Fact or tions to the audience sharing false claims. The gap

A woman holds a sign with an image of then- presidential candidate Jair Bolsonaro that reads: “He lies on WhatsApp.” Complex political divisions and the spread of disinformation and misinformation in Brazil were factors during Brazil’s elections in 2018.

18 REUTERS/Nacho Doce brazil

between audiences consuming disinformation and LOOKING AHEAD: BRAZIL those sharing verified content can in part be attributed to Brazilians increasingly getting their political information from WhatsApp, rather than from independent Brazil’s 2018 election cycle exposed the array of chal- news websites. Although various Brazilian fact-check- lenges disinformation and misinformation can pose ing organizations asked users to send the organiza- for legitimate, fact-based political discourse, and tions messages they had received on WhatsApp they the effects inauthentic online behavior can have in suspected could be false, many of the organizations decreasing public trust in institutions. The election debunking disinformation did not publish their find- cycle also showed disinformation can have the power ings directly on WhatsApp, but rather on their web- to set the agenda for political debates and news cov- sites or social media. This gap also existed on tradi- erage. In 2020, Brazil will elect representatives in more tional social media platforms. Atlantic Council findings than 5,000 municipalities—disinformation and misin- showed that when Bolsonaro was stabbed, for exam- formation are likely to emerge. ple, debunking efforts did not reach the same groups The next two years will lay the groundwork for stron- that shared inaccurate information about the incident ger digital resilience in Brazil. In 2019, Brazil’s Supreme on Facebook.79 Court will rule on two cases that can set import- Another major challenge for these organizations was ant precedents in the mobile-first country regarding assuring they were not further amplifying disinforma- encrypted messaging platforms. Some civil-society tion, as they risked sending verified information to organizations hope the court’s decisions will support people who had not heard the rumor in the first place. encryption and Brazilians’ rights to privacy. Acknowledging the importance of encryption, Civil-society Organizations and nonetheless, does not preclude technology and social media companies from continuing to recognize and Academic Institutions address the ways their platforms have been and could ivil society and academic institutions were piv- be used by illicit actors as vectors for the spread of dis- otal in the fight against disinformation during information. Technology and social media companies, Cthe Brazilian elections. The Eleições Sem Fake as well as encrypted messaging platforms, must con- project for example, led by the Federal University of tinue working with fact-checking organizations, the Minas Gerais (UFMG), was an initiative that created a media, and electoral authorities to prevent the spread way for media to have a glimpse into what was being of disinformation in Brazil’s next election cycle. shared within public groups on WhatsApp. Using Brazilians will also look to media to adapt based on automated scripts, the UFMG team identified and lessons learned from 2018. Since the elections, inde- monitored 350 public groups on WhatsApp. Eleições pendent news outlets have moved in a promising Sem Fake gave data access to fact-checkers and the direction, not only by continuing to debunk false infor- media, but did not publish personally identifiable mation but also by focusing on trying to explain disin- information about group members. Nevertheless, formation. The Bolsonaro administration’s comments WhatsApp expressed concern about researchers join- about the legitimacy and independence of media, ing public groups without explicitly identifying them- however, may impact the level of trust Brazilians place selves as researchers. Many think tanks and NGOs also on professional news organizations, and could affect analyzed Facebook, Twitter, and WhatsApp, while how verifications of disinformation and misinforma- others tackled freedom of expression discussions or tion are perceived by the population in the next elec- approached disinformation from the legal side. ITSRio tion cycle. was especially important in finding evidence of auto- Finally, the emergence of hyperpartisan media—with mation on WhatsApp groups. fringe outlets acting to spread hyperpartisan, opinion-based viewpoints, rather than taking on the role of independent watchdogs—has the potential to nega- tively shape the way citizens engage with politics. With Brazilians more aware of the existence of disinformation and misinformation, stakeholders from electoral bodies to technology and social media companies should continue working to address these challenges. On an individual level, voters can continue to question and verify narratives and to strengthen their digital resilience from the moment they get online.

19 disinformation in democracies

DISINFORMATION IN THE 2018 ELECTIONS

MEXICO By Maria Fernanda Pérez Argüello and Donara Barojan

key findings Disinformation in Mexico’s The hiring of commercial Verificado 2018, in uniting elections comprised bots for financial gain, a network of journalists, primarily automation and the use of political bots civil society, and other artificial amplification, for the spread of specific organizations under rather than organic electoral messages, and a central hub and one disinformation. disinformation about the single brand, generated electoral process at the significant impact in state level stood out as promoting transparency the three most worrisome and accountability. The trends of disinformation model should be replicated around Mexico’s elections. ahead of other elections.

CONTEXT political establishment. As of February 1, 2019, he had On July 1, 2018, Andrés Manuel López Obrador an approval rating of 86 percent.81 (AMLO) was elected president of Mexico after forging The 2018 elections were also the biggest in Mexico’s the “Together We Will Make History” coalition, com- history. Over 18,000 seats—629 at the federal level posed of the new National Regeneration Movement and more than 17,500 at subnational levels, including (MORENA) party, the Labor Party (PT), and the Social nine governorships—were up for grabs.82 The elec- Encounter Party (PES). AMLO’s 2018 victory was both tions were also the most violent: electoral authorities historic and record-breaking—historic for its defeat in Mexico registered a total of 774 aggressions against of the traditional political parties, and record-break- politicians. One hundred and fifty-two politicians ing because it was the first time a president received were murdered, 48 of which were pre-candidates almost 54 percent of votes, attributed to record-high and candidates running for office.83 According to the levels of turnout. More than 59 percent of Mexico’s Specialized Attorney for the Attention of Electoral eligible electorate came out to vote.80 AMLO, who Crimes (FEPADE), between January and September served as mayor of Mexico City from 2000 to 2005, 2018, law enforcement initiated over 1,062 investiga- had run for president in 2006 and again in 2012. With tions of election-related crimes.84 Guaranteeing the an unprecedented lead over the runner-up candi- safety and security of candidates ahead of future date Ricardo Anaya from the National Action Party elections will be key to assuring the integrity and (PAN), López Obrador’s resounding victory shook the legitimacy of future votes.

20 mexico

Supporters of Mexico’s then-president-elect Andrés Manuel López Obrador (commonly referred to as AMLO) stand in front of the National Palace, a day before López Obrador was sworn in as president of Mexico.

Modest Growth Rates, Deterioration journalists, and twelve deaths directly related to of the Rule of Law, and Changing journalistic activities.89 The job of the media remains Trade Environment dangerous in Mexico—as of February 22, 2019, a total of fourteen activists had been killed from the ears of modest economic, a deterioration in time AMLO took office.90 the rule of law and individual security, and AMLO’s victory can be attributed to a successful Yuncertainty around the fate of the North campaign aimed at resolving these systemic prob- American Free Trade Agreement (NAFTA) rene- lems plaguing the country. His focus on Mexico’s gotiations during the campaign cycle created new rampant corruption and rising crime rates resounded openings for an outsider to enter the political fray.85 with millions of disenchanted voters. But AMLO can In the span of two years, from 2015 to 2017, Mexico also attribute his win to his campaign’s adept use dropped forty places in Transparency International’s of social media. Presidential candidates allocated Corruption Perception Index.86 In Latin America, approximately 25 percent of their budgets to online the same index ranked Mexico higher than only outreach, up from only 5 percent six years before.91 Guatemala, Nicaragua, Venezuela, and Haiti.87 AMLO spent less money when compared with his Regarding violence, the 2018 murder rate of more adversaries, spending 85 million pesos, compared than thirty-three thousand homicides was a 15-per- to Ricardo Anaya’s 338 million and Jose Antonio cent increase from 2017, and constituted the highest Meade’s 302 million. 9293 Even with the disparity in homicide rate in modern Mexican history. Mexico is money spent, AMLO dominated the conversation on also the second-most dangerous country for jour- social media platforms like Twitter, where 40 percent nalists, second only to Syria in 2017.88 In 2017, Article of all Tweets about the election mentioned him, while 19 registered 507 aggressions against media and his rivals were under the 20-percent mark.94

REUTERS/Daniel Becerril 21 INFORMATION AND source for news. 101 As such, the Atlantic Council found ELECTION ECOSYSTEMS that online actors invested in creating and leasing botnets to amplify political messages. Whereas in Brazil disinformation was mostly organic in nature, in Mexico automation and artificial ampli- Mexico’s Election Ecosystem fication of messages permeated information channels during the 2018 elections. To understand why oting in Mexico is mandatory.102 Anyone over this was the case, it is important to acknowledge how the age of 18, with the exception of felons, Mexicans consume information, as well as how the Vcan vote in Mexican elections. Every six years, country’s electoral system functions. Mexicans vote for president and elect the 128 senators and 500 members in the Chamber of Deputies. Mexico’s Information Ecosystem Midterms for the Chamber of Deputies take place every three years. Constituents vote directly for their n a 2018 Reuters Institute survey, 90 percent of candidate, and the candidate with the highest num- respondents chose “online” as their go-to source ber of votes wins. Mexico’s nine predominant political for news. 95,96 The same report finds that 62 per- parties are primarily financed through public funding, I 103 cent of respondents consume news through televi- following a specific allotment procedure. To some sion, and 71 percent via social media, although trust extent, this funding structure helps create a level in social media is low, at only 40 percent.97 In Mexico, playing field; in addition, private funding may not there are over 80 million active Facebook users, and exceed public funds. Effective coordination between Facebook is the top social media platform used for national and local electoral authorities in Mexico is no news consumption (61 percent of respondents say easy task. Since 2014, the Instituto Nacional Electoral they use Facebook). 98,99 (INE) has been responsible for organizing elections As in past elections, social media played an import- by harmonizing standards and processes for local ant role in pre-electoral conversations and was there- and federal votes, guaranteeing the electoral rights fore a natural target of digital manipulation. Unlike of citizens, and counting votes at the national level. other Latin and South American countries, Mexico 104105 The Federal Electoral Tribunal (TEPJF), created has one of the highest Twitter penetration rates in the in 1990, is the highest authority on electoral mat- world, which in 2018 stood at around 49 percent.100 ters.106 These are considered the “cornerstones of The Reuters Digital News Report found that 23 per- the Mexican electoral ecosystem,”107 involved in all cent of the Mexican population relied on Twitter as a aspects of elections.

Supporters of Ricardo Anaya, then-presidential candidate for the National Action Party (PAN), take photos using their cell phones during his closing campaign rally at the Angel of Independence monument in Mexico City in 2018. Voters in Latin America increasingly rely on mobile phones to engage around elections.

22 REUTERS/Alexandre Meneghini mexico

The Atlantic Council investigated his claims and DISINFORMATION AND demonstrated that Victory Lab was boosted by apparently commercial bots—bots whose activity MISINFORMATION IN can be rented out to anyone willing to pay for likes or reposts—from East Asia on Twitter, while its Facebook THE 2018 ELECTIONS posts were liked by hundreds of accounts from South Asia and Brazil. he Atlantic Council’s #ElectionWatch Latin Partisan political messages during the election in America effort in Mexico was conducted in Mexico showed a similar pattern of amplification. This Tpartnership with media outlet Animal Politico does not definitively tie them to Victory Lab, but the and fact-checking coalition Verificado 2018. A coa- mass liking of Mexican political posts by Indian and lition of over eighty international, national, and local Brazilian accounts indicates false amplification pur- media organizations, civil-society organizations, chased online.111 research groups, and private companies, Verificado On Twitter, the Atlantic Council also observed 2018 was born after the September 2017 earth- amplification of political posts in Mexico by com- quake in Mexico. It was inspired by Verificado19S, mercial bots that appeared to come from Russian- the collaborative initiative that came together speaking accounts. The amplification was on a small within a context of disinformation around the earth- scale, in the low hundreds of interventions. The quake to fill information gaps left by the govern- accounts posting these retweets shared content in a ment, to counter false information, and to connect wide range of languages and themes, indicating that citizens who needed help with those who were they were most likely commercial bots for hire.112 able to provide it. 108 Verificado 2018 was created Despite the Russian origin, there was no evidence to to address false information around Mexico’s elec- suggest that these were linked to a Kremlin influence tion—the project was active throughout the elec- operation, despite concerns and public statements toral cycle and became a national entity that had to from national security officials in the United States.113 be responded to as it uncovered cases of disinfor- Atlantic Council research found that interventions mation and automation. were connected to commercial accounts, and on too In Mexico, while the spread of disinformation by small a scale to suggest a state-backed campaign. several actors—constituents, specialized media This is more likely to have been a case of an individual firms, presidential and local candidates, and others— buying amplification online. certainly exploited existing political rifts and insecu- rities to amplify distorted narratives, disinformation Bots Amplifying Political Content around the election was marked most prominently by automation. Disinformation comprised manu- he Atlantic Council also uncovered examples of factured, artificial dissemination that included both bots amplifying narratives to impact election coordinated and automated amplification of narra- Tresults. Some political parties utilized political tives online. Particularly, automation comprised the bots—bots created to amplify one political message use of bots and bot networks to artificially spread or stance. messages. One such case showed the use of political bots in the state of Puebla, where a small yet hyperactive Artificial Amplification of Content for network of bots was amplifying content from two PRI party candidates running to represent the state Commercial Gain of Puebla—PRI senate candidate Juan Carlos Lastiri n Mexico, the Atlantic Council uncovered cases of Quiros and PRI gubernatorial candidate Enrique a “diseconomy” of buying and selling social media Doger.114 Unlike commercial bots, which promote a Iengagements—likes, shares, and comments—for variety of brands and services in addition to politi- commercial gain. Researchers uncovered the use of cians, the Puebla bots promoted the two candidates the global market for false amplification to promote and the PRI party’s campaign materials exclusively. political or partisan content. On Twitter, this took The bot accounts were created between May 8 and the form of automated bots;109 on Facebook, it con- May 16, 2018, and had their location set to the state of sisted of commercial groups that coordinated large- Puebla. This was the first indicator the accounts were scale responses to posts in return for payment. political, not commercial (a commercial bot can be The most egregious case concerned a Mexican used for political purposes, but a political bot is cre- entrepreneur, Carlos Merlo of Victory Lab, who ated for the sole amplification of political messages), claimed to control a network of millions of fake as political bot herders tend to assign the bots loca- accounts on Twitter and Facebook. He claimed he tions where the candidate, or the party they are pro- used these networks to serve politicians and busi- moting is running to guarantee the content promoted nesses for substantial sums (up to the hundreds of by the bot accounts trends within that constituency. thousands of dollars).110 The bots used by the PRI party in Puebla were

23 disinformation in democracies

relatively sophisticated. They operated in small numbers and refrained from posting repetitive posts to avoid Twitter’s spam detection. While the bot activity created a false sense of engagement, it failed to influence the election outcome, as neither of the two candidates won their respective seats. The bot activity continued until election day.115 The Atlantic Council discovered that three days before the election, pro- and anti-AMLO bots were arguing on Twitter under a pro-AMLO hashtag #AMLOFest. The anti-AMLO bots amplified a video called “AMLO’s Dark Secret,” which falsely accused the candidate of murdering his own brother and best friend. The bots on both sides were highly active, with as little as ten bots posting more than one thousand tweets in a matter of hours. The bots managed to stifle the organic conversations under the same hashtag, illustrating their broader capacity to stifle free speech, not by suppressing it, but rather by drowning it out with automated messages and activity. Bots managed to stifle the organic Disinformation Regarding the Electoral Process and Claims of Electoral Fraud conversations under ocal elections in Mexico registered high levels of vio- the same hashtag, lence and various electoral irregularities—in some illustrating their Lcases, electoral authorities nullified elections or called for vote recounts as a result. broader capacity to The gubernatorial election in the state of Puebla was the most salient case. On election day, Puebla registered high stifle free speech, levels of violence, with reports of people murdered and not by suppressing ballots stolen or set on fire. Following a close race between the two leading candidates for governor, electoral author- it, but rather by ities announced official election results would be delayed by several days.116 drowning it out with On that occasion, the Atlantic Council reported that automated messages bot-like users tried to influence and promote misleading information about the electoral process on Twitter. These and activity. bot-like accounts promoted a series of hashtags to encourage the victory of one candidate despite official results not having yet been made public. The hashtags promoted included #PueblaEligióPAN and #YoVotéXMarthaErika in favor of Martha Erika Alonso as the elected governor of Puebla. In response to the latter hashtag, the bot-like accounts also promoted #YoNoVotéXMarthaErika in support of Miguel Barbosa, the other contender. Two and a half months after the elections in Puebla, and with more than 500 complaints of electoral irregularities, Mexico’s highest electoral court demanded a recount of the votes.117 While the vote recount was under way in Mexico City, the Atlantic Council again observed bot-like accounts promoting the victory of one of the leading contenders over the other during the vote recount, before electoral authorities released official results. The Puebla battle on social did not stop once electoral authorities released the final results. It was further exacerbated following the death of Martha Erika Alonso and former governor Rafael Moreno Valle in a helicopter crash, when hashtags emerged on social media claiming the crash had not been accidental.

24 mexico

Verificado 2018: A Model in the Fight information in his defense. As Against Disinformation election day approached and López Obrador continued to lead, disinformation shifted to the By Tania Montalvo, Executive Editor, Animal Politico and voting process and Mexicans’ right Coordinator, Verificado 2018 to vote: how to vote, where to vote, what constituted an electoral mexico’s 2018 elections were the hundred published entries. crime, and alleged prohibitions at largest in Mexico’s history. With This number does not take into voting booths. 3,400 political posts up for grabs, account information published We cannot know for sure the digital world became a key by Verificado members, meaning how disinformation affected battleground in the fight for votes, the impact was likely much the candidates, but we know and disinformation became a greater. On AnimalPolitico.com, Verificado 2018 affected how popular tool to engage the over for example, Verificado 2018 disinformation went viral and the sixty million Internet users and posts garnered almost ten million way in which readers interacted thirty-five million WhatsApp users visits. Verificado 2018 was also with it: at the beginning of who rely on digital platforms to able to engage voters through the project, each piece of receive and exchange information. different channels. For two disinformation would garner In response to this reality, a months, the group’s WhatsApp around 150,000 shares while at group of journalists developed number brought in a total of the end, days before the election, Verificado 2018, a collaborative 9,600 subscriptions and 60,700 it was hard to spot pieces of effort that put citizens at the core interactions. disinformation that had been of electoral and digital debates. Verificado 2018 saw shared that many times. Verificado 2018 took inspiration disinformation change throughout Verificado 2018 showed that from ElectionLand in the United the electoral process as the collaboration must underpin States and CrossCheck in France, public debate shifted. At the start the fight against disinformation. both of which brought media of the campaign, for example, The initiative highlighted the together to fight disinformation. disinformation was directed at importance of carrying out the Verificado 2018 embodied a new Andrés Manuel López Obrador fight against disinformation model of collaboration. Comprised (the Movimiento Regeneración on multiple fronts, and of of a central working group that Nacional (MORENA) candidate employing different formats and later became a news agency for and leading contender) and José communication channels. more than eighty allies, Verificado Antonio Meade (the incumbent Disinformation in Mexico 2018 enabled all members to work government’s candidate). After continues. Today, it is not only under one brand that conveyed the first presidential debate, focused on politics but also on a message of unity in the fight when polls and opinion leaders social issues, and it constitutes against disinformation: a group indicated Ricardo Anaya a discourse of hate. The next of allies working to promote (the Partido Acción Nacional step will be to reinforce the informed voting. and Partido de la Revolución effort initiated by Verificado Thanks to the large network Democrática (PAN-PRD) Coalition 2018, to prevent disinformation and common brand Verificado candidate) could win the race, surrounding materials that explain 2018 created, Mexican national disinformation about him began disinformation’s consequences and local television stations, to surface while disinformation and involve citizens. It is not broadcasters, and print media, as about Meade decreased. López enough to say that a piece of well as digital media, picked up Obrador, who led in the polls information is false—we must and broadcast the group’s content throughout the campaign cycle, show the effects of disinformation nation-wide. remained the primary subject of and share with citizens the What did Verificado 2018 false news, from both opposition responsibility of creating and achieve? Verificado.mx registered groups and supporters, who consuming quality content. 5.4 million visits to its four created and disseminated false

25 disinformation in democracies

Foreign State-Funded Media Outlets Encrypted Messaging Platforms head of Mexico’s presidential elections, lthough in Mexico Atlantic Council research then US National Security Adviser H.R. focused primarily on Twitter, a platform AMcMaster warned that the Russian gov- Awith which Mexicans consume informa- ernment had launched a campaign to influence tion at high rates, messaging platforms such as the country’s elections.118 As part of the Atlantic WhatsApp remain an important source of infor- Council’s #ElectionWatch research into potential mation; according to a 2019 survey, approxi- Russian influence in the Mexican elections, the team mately 17 percent of Mexicans spend at least six looked at Russian-funded media outlets publishing hours per day on the application.120 The true scale in Spanish to determine if any were trying to influ- of disinformation campaigns on encrypted mes- ence the vote.119 The Atlantic Council analyzed the saging platforms cannot be fully measured, but coverage of the elections in Latin America across Atlantic Council research shows that content Kremlin-funded media outlets such as RT, Sputnik, shared on WhatsApp is often reflective of content and News Front, as well as the pro-Kremlin website shared on open social networks such as Facebook globalresearch.ca. Although these outlets covered and Twitter.121 In Mexico’s case, more research is the Mexican elections and had a strong anti-US needed ahead of the country’s next elections to bias, none demonstrated explicit bias toward par- determine how narratives form and are spread on ticular parties or candidates in Mexico. WhatsApp.

A man takes a picture at Zocalo Square, a day before Mexico’s then-president- elect Andrés Manuel López Obrador will be sworn in at the Congress, in Mexico City in November 2018.

26 REUTERS/Daniel Becerril mexico

IMPACT ON THE ELECTIONS

Although the influence of disinformation and automation on voter decisions is difficult to measure, the following analyses based on Atlantic Council case studies published over the course of 2018 paint a picture of the impact of disinformation and automation on Mexico’s elections.

National Level: Little Discernible Impact ost artificial amplification observed during the campaign period comprised attacks on AMLO, the winner Mof Mexico’s presidential election. AMLO’s significant victory across gender, age, and education levels—especially among highly educated (university level or up) men aged 26-35—showed it is unlikely online disinformation and automation attacking him significantly impacted the results of the election at the national level. 122 The most likely impact of disinformation and artificial amplification online was of a financial nature and affected individual politicians—Mexican media reported that entrepreneurs like Carlos Merlo and others potentially charged politicians hundreds of thousands of dollars for covert social media campaigns.123 Most artificial

State Level: Local Unrest amplification t the local level, particularly in the states in which observed during the election results were contested or had a higher num- campaign period Aber of irregularities, disinformation and inauthentic engagement efforts focused on undermining the legiti- comprised attacks macy of the electoral process and the credibility of electoral bodies. In the case of Puebla, amid a vote recount, disinfor- on AMLO, the winner mation and artificial engagement on social media reflected polarization and exacerbated existing tensions between of Mexico’s 2018 opposing political groups. This became more evident following the death of Martha Erika Alonso, when new conspiracies presidential election. about the nature of the helicopter accident emerged online.

Party Level: Political Parties and Governments tlantic Council research indicated political parties in Mexico at times turn to artificial amplification to bol- Aster support for their political campaigns and candidates. According to local journalists with whom the Atlantic Council spoke, some political marketing companies also include these types of services in their proposals to political parties and government. As was seen with the government of the state of Oaxaca in southern Mexico, governments at times resort to bot-like and artificial support to promote their image. The Atlantic Council identified farm-like accounts promoting social media pages from the Oaxaca state government—an Atlantic Council analysis of online patterns of engagement in Mexico uncovered an unusual number of South Asian accounts reacting to posts on the official Facebook page of Alejandro Murat Hinojosa, the governor of the state of Oaxaca, in southern Mexico. Similar

27 disinformation in democracies

reactions can be seen on his official state government’s As Mexico’s electoral authorities sought to address page and Guelaguetza page—an account intended to the challenges of disinformation in the elections, promote the Guelaguetza festival, an annual traditional debates emerged about the balance between regula- indigenous festival celebrated in the city of Oaxaca tion and freedom of speech. When Mexico’s Tribunal every July. Electoral del Poder Judicial de la Federación (TEPJF) The very high proportion of South Asian reac- was established, disinformation online did not exist tions is unusual for Spanish-language political con- to the same extent it does today. Reactions and rul- tent in Mexico. The proportion and number of likes— ings from the TEPJF thus need to thread a fine line to in comparison to Facebook’s other options of loves, remain in the parameters of its constitutional mandate, laughs, and angry reactions—also appeared distorted. and to assure actions to stop illicit activity online do Both features suggest that the “like” reactions were not also open doors for infringement on freedoms of obtained from South Asia as part of a transaction, expression.”128 most probably in return for payment. Nonetheless, it is Furthermore, there is still a lack of clarity on how important to state that this does not suggest the South technology and social media platforms can be used Asian likes were paid for with taxpayers’ money nor during the different phases of the electoral process, that government authorities were responsible for this. whether it be throughout the campaign process, the Additional research is needed to determine whether day of the election, or in the post-electoral phase. political parties and government actors are actively engaging in artificial promotion of their platforms and Fact-checking Agencies and the Media activities on social media. Nonetheless, open-source research suggests political parties and governments he fact-checking initiative Verificado 2018 in are making some use of these strategies. Mexico was perhaps the most successful in Tcountering disinformation and misinformation, and shedding light on automation, around elections in Latin America. Over eighty media outlets—spanning international, national, and local media—joined the INSTITUTIONAL effort to raise awareness of and fact-check false information, the biggest collaborative effort of its kind. REACTIONS Civil-society organizations and academic institutions also took part. The initiative, inspired by a crowdsourced effort to map available resources in Mexico Government and Technology and Social City following a massive earthquake in September Media Companies 2017, drew also upon crowdsourcing and volunteers to help verify information and debunk hoaxes in real head of the Mexican elections, the National time. The project received funding from Open Society Electoral Institute (INE) signed agreements Foundations, Oxfam, Facebook, Twitter, Google News Awith Facebook, Twitter, and Google to increase Lab, and Mexicans Against Corruption and Impunity, collaboration with technology and social media among others, and oversaw the Mexican information platforms.124 environment over the course of four months.129 In February 2018, INE announced an agreement with Verificado 2018 systematically documented and Facebook to promote citizen engagement during the exposed false content, something that had never hap- elections.125 As part of the agreement, Facebook com- pened at scale in Mexico’s history. The group’s unique mitted to supporting citizen participation in the 2018 model, which comprised a central hub and enabled elections and to organize election-related forums and members to work collaboratively under one brand to training sessions for journalists.126 Facebook employed debunk false news and guarantee informed voting, a variety of engagement tools to encourage citizen helped to promote confidence among voters about the participation, including an electoral megaphone tool veracity of candidate statements throughout political that invited citizens to vote on July 1 and an informed campaigns, as well as the legitimacy of election results. elector button that made available candidate informa- The group also published educational and information and details about the electoral process. tional material to help constituents understand phases Likewise, INE signed an agreement with Twitter to of the electoral process.130 broadcast the presidential electoral debates on its Whereas in other countries fact-checking organiza- Periscope platform, but explicitly noted there would tions and media worked from their own newsrooms be no exchange of information.127 Finally, INE signed an and with no central command structure, Verificado agreement with Google to build on existing collabora- 2018 became in Mexico an active body that elicited tion between both entities—the agreement was mainly responses from candidates and other political stake- to provide information to citizens and collaborate in holders. The group was also a central part of the elec- the process of building a free and informed vote. tion conversation.

28 mexico

Civil-society Organizations and Academic Institutions

ivil society and academic institutions in Mexico, the integrity of the digital information environment including Mexico Como Vamos, Think Tank New ahead of future elections in Mexico. Digital media lit- CMedia, Article 19 and others also participated eracy and the ability to adequately respond to disin- in the Verificado 2018 effort. In addition, Mexico’s formation will be key in combatting efforts that aim to civil society and academia sought to discuss and undermine the credibility of electoral bodies and the address issues of disinformation through conferences legitimacy of the electoral process. and studies leading up to election day. Academics from Universidad Iberoamericana (Iberoamerican University), the Instituto Tecnológico y de Estudios Superiores de Monterre (Monterrey Institute of Technology and Higher Education), the Centro de Investigación y Docencia en Económicas (Center for Research and Teaching in Economics) as well as international experts from the National Democratic Institute, the Institute for the Future, and the Center for Research, Transparency and Accountability took part in various conferences, alongside INE and the media, leading up to the elections.

In Mexico, LOOKING AHEAD: MEXICO disinformation It is unlikely the pace of automation in Mexico will slow before the country’s 2024 presidential election and automation cycle. In fact, disinformation and automation will likely will likely remain remain vehicles for digital exploitation over the coming months and years, not only in elections, but around vehicles for digital social issues. The 2018 election cycle in Mexico demonstrated how exploitation over bots can take control of a conversation at high rates and on a broad scale. The fight against disinformation the coming months and automation is an up-hill battle—as one instance of automation is uncovered and addressed, others and years, not only emerge. Global technological developments in Artificial in elections, but Intelligence will increasingly make identifying robotic activity at scale and distinguishing citizens from bot around social issues. accounts more difficult, with enormous implications. Verificado 2018 played an important and unique role in Mexico’s fight against disinformation and misinformation before and during the elections. In collaboration and under the umbrella of one brand with a central hub, the group brought disinformation and misinformation, and the challenges associated with these illicit activities, to the forefront of public debates. The efforts Verificado 2018 spearheaded around the last elections will become increasingly necessary. Given the complexity of disinformation, more civil society organizations, academic institutions, and government agencies are likely to join the ranks of a Verificado-like project around 2024. A multi-stakeholder approach to strengthening digital media literacy for public officials as well as for voters, in a country where online news as the go-to source is rapidly gaining market share, will be key in protecting

29 disinformation in democracies

DISINFORMATION IN THE 2018 ELECTIONS

News headlines announce the election of Iván Duque as new president of Colombia in COLOMBIA 2018. By Jose Luis Peñarredonda and Roberta Braga

key findings Disinformation in In a polarized electoral As in the case of Brazil, Colombia’s elections environment, fact- countering disinformation largely comprised organic checking organizations narratives, rather than disinformation at times and the media must work instances of disinformation, amplified by media outlets to assure verifications may be more effective and political leaders. reach the same audiences in assuring that exposed to and affected by disinformation does not set disinformation. the agenda for electoral debates and media coverage.

CONTEXT

On June 17, 2018, Colombians elected Iván Duque Márquez as their next president with 54 percent of the vote after two highly contested rounds of voting. Four months earlier, the country had elected the one hundred and eight senators and one hundred and seventy-one representatives who would serve in Congress for the next four years. In both contests, three concerns that transcended borders permeated the electoral process: the effects of disinformation and misinformation on election results; the impact of high degrees of polarization in political discourse, and the consequences of citizens’ lack of knowledge or lack of trust in the electoral process.

30 colombia

The 2016 Plebiscite for the “Yes” vote.135 This generated increased awareness about the risks misinformation and disinforma- vidence of disinformation campaigns previ- tion represented for the integrity of the 2018 presi- ously emerged around the 2016 plebiscite dential elections, and government, media, civil society Eon Colombia’s peace agreement with the and technology companies ramped up their efforts to Revolutionary Armed Forces of Colombia (FARC). create awareness of disinformation and misinforma- A worrying precedent was set at that time about tion among the public and to better understand their the potential effects of disinformation on polit- effects. ical discourse. The victory of the “No” camp, a The outcome of the plebiscite became one of the vote against the peace agreement, by little more most significant topics of political debate follow- than 50,000 ballots (0.2 percent of the total) was ing the peace deal referendum in August, and fueled partly attributed by academics, the media, and non- polarization in the country. In spite of having said that governmental organizations to a seemingly orga- “Colombian people […] will have the last word,”136 nized disinformation operation conducted through then-President Juan Manuel Santos made some WhatsApp and other social media platforms by changes to the peace deal after a brief consultation some sections of the opposing right-wing cam- with opposition parties, and Congress approved it. paign.131,132,133 In an interview days after the vote, its The Centro Democrático party, as well as some sec- manager, Juan Carlos Velez Uribe, acknowledged a tors of the Conservative Party of Colombia, continued strategy of distortion of the facts.134 to aggressively oppose the deal once implementation During that referendum, the flow of organic dis- fell to Congress, while the political center and those information through family and friend groups on on the left were largely satisfied with it moving for- WhatsApp, many of them false or out of context per ward. This situation gave way to hyper-polarization a study conducted by a researcher at the University during the presidential campaign. of Santiago de Compostela in Spain, was considered The two top candidates in the presidential election to be a contributing factor to a surprising outcome, belonged to opposing sides of the political spectrum since most polls forecasted a wide margin of victory and represented the two sides of the 2016 plebiscite

REUTERS/Luisa Gonzalez 31 disinformation in democracies

A woman walks under a banner supporting the “yes” vote during the Colombian peace plebiscite in 2016.

discussion. Iván Duque identified with the conser- institutions.”137 Absenteeism remains widespread. In the vative policies aligned with former president Álvaro runoff for the 2014 presidential election, only about 48 Uribe, while Gustavo Petro, former mayor of Bogotá, percent of qualified voters cast a ballot, and congres- championed a more left-leaning agenda. The third sional and regional elections saw even worse turnout.138 runner-up, Sergio Fajardo, led a center coalition that This lack of trust exacerbated disinformation and mis- agreed with the peace agreement but promoted more information narratives during the presidential elections. pro-open-market policies than did Petro. Electoral institutions and media often found themselves Leading candidates Duque and Petro represented caught in the middle of the crossfire of a belligerent two starkly different visions for the country. While political debate, and—despite their efforts—were lim- Duque ran on a platform underpinned by opposition ited in their ability to effectively counter false narratives to the peace agreement for its alleged leniency with and disinformation about the neutrality and reliability of the FARC’s top ranks, Petro was a keen defender of their own work. the peace deal, and advanced a “progressive” agenda. These different approaches to the way forward for peace and the country’s future created an environment that enabled the spread of disinformation around the elections. INFORMATION AND ELECTION ECOSYSTEMS Lack of Institutional Trust he 2018 election took place within a climate To understand how disinformation and misinforma- of widespread public distrust in the coun- tion spread during the 2018 elections in Colombia, it is Ttry’s institutions, particularly in the orga- important to understand not only Colombia’s informa- nizations responsible for monitoring elections. tion environment, but also the institutions responsible According to a 2017 study conducted by the Institute for managing elections. for Democracy at the Universidad de los Andes, 31 percent of Colombians are satisfied with the way Colombia’s Information Ecosystem democracy works in the country, around 24 percent trust elections in the country, and 10 percent According to official government statistics, around 75 trust political parties, “a trend that coincides with percent of Colombians used the Internet at least once in the increasing disapproval toward the operation of 2017. Of those who use the Internet, around 87.5 percent Colombian democracy and its most representative use Facebook, and 87.3 percent use WhatsApp. Of those

32 REUTERS/John Vizcaino colombia on WhatsApp, nearly 93 percent open the app daily.139 As a result, politicians increasingly use digital media Linterna Verde: and communications for campaign purposes; and some have been responsible for spreading disinfor- WhatsApp Was Not the mation. 140 In 2018, all presidential candidates had an Beginning, Nor the End active and official presence on Facebook, Twitter, and YouTube. In addition, a number of social media of Misinformation in groups, pages, or accounts supporting individual Colombia candidates—sometimes grassroots, and sometimes run by political parties and organizations—sprouted during the electoral campaign. Carlos Cortés, Co-Founder, Polls conducted by advertising agencies since 2017 Linterna Verde show that social networking services like Facebook or Twitter are the most frequently used platforms for the when it came to disinformation around consumption of news after television, with nearly half Colombia’s 2018 presidential election, of Colombians saying they use the platforms to do Linterna Verde found that encrypted so. Surveys also place the popularity of social media messaging platforms like WhatsApp above online news media, radio, and print media.141 did not have an independent or isolated While Colombian audiences increasingly rely on impact on elections. In Colombia’s social media over traditional media for information, information space, the encrypted electoral authorities and legislations have not yet messaging platform was complementary caught up to address the changing media environ- and parallel to open social networks ment. Media regulation around elections were insti- such as Facebook and Twitter. In that tuted in a pre-social media era, making it particularly sense, disinformation was not exclusive difficult for the appropriate authorities to tackle infor- to WhatsApp, and it did not mark the mation distortions online. beginning nor the end of attempts to manipulate election conversations. Colombia’s Election Ecosystem Our research did find that presidential campaigns have the capacity to undertake hree elections took place in Colombia in 2018. disinformation operations. Although we Congressional elections occurred in March, as cannot say that campaigns took steps Tdid primaries for the Centro Democrático Party to insert content onto WhatsApp, the and Petro’s Progresistas movement, which Duque campaigns’ infrastructure and operations and Petro won respectively. The first round of presi- offer powerful incentives for them to be dential elections took place on May 27, and the sec- a source of, or point for, reinforcing false ond on June 17. None of the three elections, nor the or manipulated narratives. Campaigns August 26 referendum on anti-corruption measures, have databases, community managers, generated a turnout higher than 53.7 percent.142 designers and –of course– computers, all Voting is not mandatory in Colombia but is incentiv- working to implement communication ized by benefits that include discounts on requests for strategies. In that regard, and as a couple expedition of passports and public university tuition. of sources told Linterna Verde, although The main institutions in charge of organizing and those assets are used for legitimate guaranteeing the integrity of elections in Colombia actions, they can, at times, also be used to are the National Registry of Civil Status (RNEC, advance disinformation on social media. Spanish acronym) and the National Electoral Council WhatsApp’s usage in Colombia has (CNE, Spanish acronym). evolved with social practices. Linterna The RNEC issues national identification cards to all Verde’s research found that voters Colombian citizens and maintains the “electoral cen- increasingly tend to reject unsolicited sus,” a list of all citizens able to vote: everyone over 18, political campaigning and activism on except for convicted felons, enlisted members of the WhatsApp. In that sense, successful military, and the police. The RNEC is also responsible attempts to manipulate and disinform for setting up polling stations throughout the coun- require articulated operations on the try and in Colombian consulates worldwide, recruit- ground. With such influence, WhatsApp ing electoral jurors for polling places, and conduct- can indeed reinforce environments for ing a first count of the results after the election. The manipulation and radicalization. CNE is the highest electoral tribunal in Colombia. It has the last word on election results and oversees investigating and sanctioning infractions to electoral jurisdiction.

33 disinformation in democracies

Colombian presidential candidates German Vargas Lleras, Gustavo Petro, Sergio Fajardo, Iván Duque and Humberto de la Calle take part in a presidential debate at El Tiempo newspaper in Bogotá, Colombia in May 2018.

There are eleven “electoral offenses” in the have to be carefully measured against its potential Colombian criminal jurisdiction, including fraud, voter consequences for democratic liberties such as free- impersonation, and disturbance of the elections. dom of speech or privacy. Balancing between these “Electoral coercion,” or making someone vote in a two imperatives is perhaps the biggest challenge for given way by means of “a deceitful maneuver” is pun- legislators and media, and for regulators in Colombia ishable by at least four years in prison. They are prose- and elsewhere. cuted by the Office of the Attorney General and ruled on by criminal judges. Several laws regulate media and political communication during elections. Candidates must declare all media and advertising expenditures, as these are DISINFORMATION AND considered campaign expenses and therefore count against spending limits. Political advertising in media MISINFORMATION IN and on the streets, polling, and public meetings are THE 2018 ELECTIONS banned for two days before the election. Colombia’s regulatory framework around elections, while robust, remains insufficient for addressing the To explore how disinformation played into public dis- challenges a changing digital communication envi- course and impacted electoral outcomes in Colombia, ronment can pose to the integrity of an election pro- the Atlantic Council’s Digital Forensic Research Lab cess. Laws currently in place do not prevent unofficial and Adrienne Arsht Latin America Center conducted political campaigning that happens on social media, open-source investigations and published articles nor polling conducted by social media users. Since explaining cases of disinformation and misinforma- the framework requires that only campaigns declare tion in real time. expenses; individual users have no barriers or limits to Disinformation and misinformation on social media unofficially paying for and disseminating advertising were common throughout the 2018 election. During messages in support of a candidate online. the presidential campaign period (March 12 to June The regulation criminalizing coercion also seems 16), Colombian fact-checking organizations received insufficient for prosecuting online disinformation daily requests to verify allegedly false claims.144 and misinformation, as it requires that the prosecu- In Colombia’s case, polarization seemed like the tor prove a person voted as they did because of false biggest driver of disinformation and misinformation information produced by the defendant. In this sense, during the presidential campaign. #ElectionWatch the difficulty in demonstrating that someone would research in Colombia—done in partnership with La have voted differently if they had not been in contact Silla Vacía, Linterna Verde, and Centro Gabo—showed with disinformation or misinformation online is a sig- the prominence of organic disinformation and mis- nificant obstacle in prosecuting such crimes.143 information in Colombia, fueled by polarization and However, any strengthening of this regulation would often amplified by high-level political representatives,

34 REUTERS/Henry Romero colombia and at times news outlets. inconsistences are corrected at a later stage. Hence, At the same time, a handful of new, highly parti- the fraud claim was based on a minority of tran- san media outlets, Facebook groups and pages, and scripts that were going to be corrected. Colombian Twitter profiles, drove significant engagement and NGO Electoral Observation Mission (MOE being the traffic by catering to audiences engaged on both ends acronym in Spanish) verified this through an inde- of the political spectrum, often by offering partial or pendent study, which showed only 2.8 percent of blatantly false accounts of the news. Atlantic Council the forms had apparent errors.150 #ElectionWatch Latin America findings,145 as well as Another instance of disinformation amplified by other researchers’ results, also indicated disinforma- politicians happened after Colombians attending tion circulated prominently on the messaging app a Duque campaign rally held by former President WhatsApp146, one of the primary tools for communi- Uribe were attacked by a swarm of bees. Congress cation in the region. members close to Uribe amplified an unsubstanti- ated claim that Petro followers were responsible for Polarization and Amplification of the incident— while police reports showed the bees had been disturbed by the sounds of Álvaro Uribe’s Disinformation by Elected Officials helicopter. The false claims, amplified on the social hrough open-source investigations conducted media channels of candidates and their support- before, during, and immediately following ers, went viral, and like in many other cases, official Tthe two rounds of elections between May and reports rarely reached the same audiences as those August 2018, the Atlantic Council found that verified originally exposed to the disinformation. and non-verified social media accounts of political #ElectionWatch Latin America case studies leaders, candidates and political parties of all politi- showed that content circulated along political lines cal leanings were among the main amplifiers for the in Colombia and proved those who shared or saw a spread of false or misleading content online. false claim were rarely also exposed to verified and #ElectionWatch case studies published during the accurate information about the same issue on social election cycle consistently showed political lead- media. Larger-scale studies confirmed this insight–a ers and their online accounts as major sources and report from MOE based on an analysis of more than amplifiers of disinformation. They directly started147 forty-four million social media publications about or helped distribute148 false or misleading claims that Colombian elections also concluded that “when advanced or solidified their positions. In one instance, information is refuted, it is highly unlikely that [the Petro’s social media accounts amplified an accusa- debunking] gets to the community that viralized it tion of electoral fraud against him during the first in the first place.”151 round, when some voters claimed amendments and corrections had been made to E-14 forms, the docu- Disinformation and Misinformation ments on which polling officials write how many votes each candidate had at each polling station. This piece Shared by Hyperpartisan Media and of disinformation was amplified after the candidate Blogs retweeted the claim to his followers. alse claims not only spread faster and more The claims were not of a scale large enough to be widely than verifications, it did so in spaces substantial and were based on a misunderstanding Fwhere mostly partisan content circulated. This about the electoral process. Still, they spread very was another consequence of the intense polariza- quickly on social media, in part because they played tion Colombia experienced during these elections, along a narrative of mistrust of authorities that suited as one-sided political pages and groups amassed Petro’s anti-establishment rhetoric. He used the epi- large followings on social media and managed to sode to push his campaign: he briefly called for a pro- create significant communities around them— test “if there was evidence of a fraud.”149 some with more than one million fans. These digital E-14 forms are filled out manually and are the first spaces often worked as “echo chambers” for con- step in a complicated vote count that lasts several tent that favored one side and attacked the other, days before results are made official by the National including false and misleading claims that went Electoral Council, Colombia’s supreme electoral largely unchecked. authority. The forms are scanned and published online There were also several hyperpartisan websites and are freely accessible for the public to scrutinize. that mirrored news outlets but published content In this case, the National Registry explained that the with various degrees of distortion during the elec- main issue stemmed from the existence of three dif- tions. Some, like El Nodo or Voces.com.co, were ferent E-14 forms: one used to transmit preliminary political blogs that carelessly combined opinions results on election day, another scanned and pub- and facts in their pieces, pushed a one-sided agenda lished on the National Registry website, and another outside of the standards of objective journalism, and used for the official vote count. The process pro- claimed to be “independent media.” Others, like El vides for handwriting mistakes in the E-14 forms, and Expediente, used some journalistic techniques but

35 disinformation in democracies

TABLE 5 Top Ten Media Outlets for Social Media Engagement in Colombia Hyperpartisan media Gustavo Petro Outlet Iván Duque

publimetro.co 198,531

pulzo.com 446,724

835,805 elespectador.com 735,124

302,038 eltiempo.com 203,164

191,249 lafm.com.co 437,520

250,633 rcnradio.com 219,881

1,039,581 semana.com 638,419

736,637 voces.com.co 1,065,793

275,604 wradio.com.co 346,818

557,305 youtube.com 280,557

385,376 noticias.caracoltv.com

193,597 oiganoticias.com

1.0 0.8 0.6 0.4 0.2 0 0 0.2 0.4 0.6 0,8 1.0

SOURCE: Buzzsumo.

consistently published unbalanced and inaccurate everyday communication tool for many Colombians. reports, while others, like Oiga Noticias, simply spread WhatsApp’s encrypted design makes gathering false news. complete information about the content circulating Political leaders on both sides shared these web- on the platform very difficult. The messaging plat- sites’ contents, and some of these outlets benefited form began working with fact-checkers and, for the from the existing infrastructure of Facebook political first time, gave a journalism non-profit the ability groups that gave them a sizeable audience of like- to receive questions from the public during Brazil’s minded users. Voces.com.co, for example, became elections.153 Likewise, researchers like those at the one of the most-shared sources for political content University of Minas Gerais’ Monitor Do WhatsApp on Facebook during the campaign. Despite being less created tools to partially track what happens on the than a year old, its posts had more interactions during platform.154 However, both tools appeared after the the electoral campaign than posts from most tradi- Colombian elections and were tailored to the Brazilian tional media in Colombia.152 ecosystem. This made any open-source analysis about disinformation during the Colombian elections neces- WhatsApp as a Vector for sarily incomplete. However, working with La Silla Vacía, a news website Disinformation and local partner in Colombia, #ElectionWatch Latin hatsApp was a vector for the spread of America results showed how disinformation and misinformation—and disinformation—during information spread on WhatsApp was cross-posted Wthe electoral cycle in Colombia, as in Brazil. to ‘open’ social media platforms by users. During the The messaging app is ubiquitous in Colombia; it is elections, La Silla Vacía ran a fact-checking service offered by all carriers on a zero-rating basis—which where users forwarded messages they received on means data consumed on the platform does not WhatsApp to the newsroom. count against overall data usage within an individ- In one example a week before the first round of ual’s mobile device service plan—making it a free the presidential elections, the outlet was sent two

36 colombia

A WhatsApp-Reliance Jio representative displays key chains with the logo of WhatsApp for distribution during a drive by the two companies to educate users in India.

SOURCE: Buzzsumo.

messages: the mayors of two different cities were research did not show that bots, or other automated being accused of pressuring school officials to vote inauthentic activity, played an influential role in for a candidate. Each message had a different city information flows during Colombia’s elections.156,157,158 and a different candidate, and one had a spelling Some evidence of behavior consistent with bot mistake the other did not. The wording was other- operation were detected, including posting rates of wise almost identical, and neither story could be ver- hundreds of tweets per hour or periods of intense ified byLa Silla Vacía journalists. publication activity after months of dormant activ- The Atlantic Council could verify that both ver- ity. However, this activity had limited influence, since sions of the messages were being shared on par- it was conducted by accounts with few followers or tisan political groups by influential users, and gar- interactions.159 nered significant engagement there. The extent to which WhatsApp messages and content were re-cir- Narratives culated on ‘open’ social media platforms is beyond the scope of the open-source methodologies and isinformation in Colombia spread along par- research, but anecdotical evidence suggests it was tisan lines–while Duque was constantly common practice. Still, Linterna Verde’s analysis Dattacked for his alleged proximity to a sup- of the role of WhatsApp in Colombian elections posedly corrupt establishment, Petro was hit by showed how these so-called “chains”—political mes- false information that portrayed him as a left- sages that are created to be shared in groups—were wing terrorist close the Nicolás Maduro regime in often created by political campaigns and reached Venezuela, and as an outlier who did not conform users through users who often volunteered on cam- with societal values. paigns and shared the messages with their friends The chart on page 39 shows some of the major and families.155 narratives against each candidate. Unlike what happened in other parts of the world— such as Mexico, Pakistan, Italy, or Indonesia—the

REUTERS/Rupak De Chowdhuri 37 disinformation in democracies

La Silla Vacía: In Colombia, Fact-Checkers Got Checked

Juan Esteban Lewin, Editor-in-Chief, evidence that the messages had La Silla Vacía been shared within WhatsApp groups. La Silla Vacía’s commitment was to fact check in colombia, the strength and disinformation explode in these messages and send verified spread of disinformation was not public debates around the information back to users, while disproportionate to those in other 2016 referendum on the peace users committed to circulating countries. Although determining agreement—it was when false debunked content back into their how disinformation shaped information began to circulate WhatsApp groups. election outcomes is difficult to widely on WhatsApp and This was both a success and a do with certainty, disinformation politicians and voters began failure: the effort was a success became part of political discourse sharing and amplifying these because the fact-checking and the public debate. During messages that fact-checking took initiative became very popular Colombia’s 2018 elections, some off in Colombia. among users; the effort was a politicians attacked both the These efforts were largely in failure because the fact-checking media and their opponents’ vain. Fact-checkers debunked team could not keep up with arguments by accusing them of but did not prevent the spread demand. spreading “fake news.” of false narratives claiming, for Disinformation continued At the same time, political example, that pensions would be to spread in the following divides in Colombia shifted: the reduced to pay monthly wages months and years. During the divide between left and right for demobilized guerrillas or that 2018 elections, the spread of became starker in a country the Colombian peace agreement disinformation became a clear where political divides have included “gender ideology” problem as the left-right divide historically not been as distinct that would influence the sexual deepened and people on both as in the rest of Latin America. orientation of children. sides repeated disinformation and The division was exacerbated La Silla Vacía quickly came to misinformation that reinforced during the campaigns around the the conclusion that fact-checking their points of view. 2016 plebiscite, when Colombians can only help stop hoaxes when Although La Silla Vacía simply were asked to vote for or against verified information is distributed did not have the resources a peace agreement between the by real people in whom necessary to repeat the Colombian government and the Colombians trust. WhatsApp experiment, the team Revolutionary Armed Forces of To address this challenge, La fact-checked through Twitter, on Colombia (FARC). It was in this Silla Vacía pioneered an effort television, and on our website— markedly polarized campaign that to verify messages circulating the pieces we published verifying disinformation took hold. on WhatsApp and to distribute information received a great deal La Silla Vacía began fact- verified content also through the of both attention and criticism. checking public discourse in messaging platform. We asked Perhaps that’s the future: an 2014 through “El Detector” users to send in content they environment where fact-checkers (The Polygraph). We saw assumed was false, along with are fact-checked.

38 colombia

Narrative Background Examples of False Information Petro, his circle, and his During his time as Bogotá mayor Uribe supporters accused supporters as vandals. (2011-2015), Petro was close to Petro of attacking the former political minorities and grassroot president with bees; false movements that some on the reports of one of Petro’s sons right identify as “vandals.” being in debt.160 Duque as the candidate of a Duque was seen as the Petro supporters aired corrupt establishment candidate favored by powerful unjustified fraud claims economic interests and against the National Registry; Colombian political elites. Duque was falsely accused of plotting fraud with a cousin who supposedly worked in the National Registry. “The FARC candidate” and the After more than fifty years as A photoshopped image of FARC Venezuela threat an armed group, the Alternative leaders wearing Petro t-shirts Revolutionary Force of the made the rounds on WhatsApp. Commons—the political party The idea that electing Petro born after the FARC signed would “turn Colombia into the peace agreement ̶ went another Venezuela” fed many through their first election cycle memes and social media as a legitimate party. Meanwhile, messages, some originating old videos showing Petro’s from Venezuela. admiration for late Venezuelan leader Hugo Chávez surfaced.

candidates’ agendas. Candidates often had to correct IMPACT ON THE disinformation or misinformation against them, and ELECTIONS media reported and covered both the falsehoods and the corrections, sometimes extensively.162 Finally, it is likely that news outlets played a role in The most pressing question about the impact of dis- amplifying some false claims through extensive cov- information and misinformation in Colombia’s elec- erage.163 An #ElectionWatch Latin America analysis tions is the extent to which they influenced voting showed that a viral Facebook post, using a picture of results. This is a remarkably hard question to answer: former adult film actress Mia Khalifa, claimed Petro each person’s decision about voting is affected by had an illegitimate daughter who announced her sup- many different variables, making it difficult to isolate port for Duque. The story was false, but some voters the effects of disinformation.161 However, four tangi- believed it even after most Colombian media reported ble effects were clear: polarization and disinforma- this. Some stories gathered thousands of interactions tion were mutually reinforcing, disinformation had an on social media, as Figure 2 shows. agenda-setting effect in news coverage, media cover- Another possible effect of disinformation in the age of disinformation and misinformation sometimes Colombian election was its likely contribution to the seemed to amplify such content, and disinformation erosion of trust in electoral institutions. Some of the had a connection with the erosion of trust in demo- biggest instances of disinformation, like the alleged cratic institutions. electoral fraud case, were based on the premise of While polarization shaped disinformation flows on suspected institutional failures or corruption. The social media, disinformation may also have deepened wide circulation of disinformation about electoral and expanded polarization. The fact that hoaxes and fraud showed the degree of distrust many Colombians corrections were shared by different communities felt toward their electoral authorities, and its wide dif- according to their political leanings—and that polar- fusion also likely deepened citizens’ skepticism and ized media were able to amass big audiences in short confusion. timeframes—show that disinformation might have Fortunately, the E-14 episode did not compromise been instrumental in engaging people at both ends of the integrity of the elections once the fraud claims the political spectrum and furthering divisions. were debunked, as Petro ended up accepting the Disinformation and misinformation also shaped results. But this episode shows there is a need for media coverage of the campaign, and of the greater understanding by the public of the intricacies

39 disinformation in democracies

TABLE 6 Media Coverage of The False Story about Gustavo Petro’s Alleged Daughter

Total Headlines Outlet Engagement

La supuesta hija no reconocida de Gustavo Petro semana.com 29,600

Supuesta hija no reconocida por Petro es Mía Kalifa rcnradio.com 14,300

Mía Khalifa, la actriz porno a la que señalaron de ser la hija negada de Petro semana.com 11,000

Petro rechaza difusión de imagen de Mia Khalifa en la que dicen que es hija suya pulzo.com 5,900

Noticias falsas sobre Gustavo Petro: dicen que esta reconocida exactriz porno es su hija pubimetro.co 500

SOURCE: Buzzsumo.

of the electoral system. It also shows official institu- video imitating the aesthetics of the Duque campaign tions might benefit from a more aggressive commu- prompted supporters to stay home and refrain from nication strategy and faster internal processes. While voting in the second round. The author was located a rumor may be amplified significantly in a matter of by the General Attorney’s Office and asked to give a minutes, an official entity often takes hours to draft, statement. approve and publish a press release with the accu- Still, government authorities had to deal with the rate information. complications of a regulation meant to balance efforts However, #ElectionWatch Latin America work, as to combat disinformation with the constitutional duty well as other institutions’ research on disinformation of protecting democratic freedoms. Also, internal-com- in social media, undoubtedly had a positive impact munication processes proved slow in effectively coun- on the Colombian information ecosystem. The countering hoaxes on social media. While misinformation try saw a sprout of events, reports, and projects that sometimes became very popular in a matter of min- aimed to approach the issue from different perspec- utes, the process of producing, approving, and com- tives. Linterna Verde’s work offers some encouraging municating an appropriate response took hours. evidence that Colombian social media users increas- 164 ingly mistrust reports shared with them online. Technology and Social Media Companies acebook partially funded several projects that sought to tackle disinformation in the coun- INSTITUTIONAL Ftry, including #ElectionWatch Latin America, Colombia Check, La Silla Vacía and El Poder de Elegir. REACTIONS These partnerships were part of global efforts by the company to identify and combat disinformation and misinformation on its platform. Facebook also ran Government campaigns in local media aimed at teaching users his was the first election in which authorities how to detect and verify misinformation circulat- identified disinformation as a risk and took ing on social media. Twitter, in turn, had contact with Tactions to proactively counter it. While the MOE and took part in communicating their results in National Registry often released corrections of false a workshop in Bogotá, though the company took no information about the election process, the Ministry part in the development or funding of MOE’s study. of Interior produced content warning the popula- Analysts have observed that Latin America and other tion about disinformation more broadly. In turn, the parts of the world do not play a proportionate role in judiciary prosecuted a landmark case, in which a the design of technology products used in the region.

40 colombia

There is the generalized concern among Colombian Red, led by other organizations, that sought to fur- observers that, for all the presence social media giants ther understand disinformation and to coordinate have in Latin American countries and their dialogue responses and actions. However, their efforts could with local journalists and disinformation research- benefit from better communication, as not all actors ers, the local populations’ concerns exert very lit- know what their peers are doing and how they can tle influence on decisions by technology companies. collaborate with each other. Cooperation or synergies Technology and communication platforms don’t have remain, therefore, limited. strategies to tackle misinformation and expedite content moderation as specific to Colombia and other Latin American countries as they do to other parts of the world.165 LOOKING AHEAD: Fact-checking Agencies and the Media COLOMBIA any media outlets in Colombia took fact-checking seriously for the first time around the 2018 As Colombia prepares for local and departmental Mpresidential elections. El Tiempo and Semana, (state-level) elections in 2019, and for presidential Colombia’s more influential print outlets, created elections in 2022, three main concerns regarding dis- dedicated fact-checking sectors or teams. Media information are likely to emerge. who already focused on these techniques, such as First, the existing infrastructure of partisan social Colombia Check or La Silla Vacía, reached a larger media profiles, groups, and pages remain opera- audience and became more influential as they made tional—these profiles, groups, and pages are likely to SOURCE: Buzzsumo. alliances with bigger and more traditional outlets. become increasingly active ahead of the elections. However, disinformation was so abundant that media Such partisan actors are likely to once again drive had a hard time deciding what to debunk. This over- false and misleading information about the candi- rode their capacity to analyze and verify rumors or dates and the electoral process. In the coming years, false information. As a result, many false statements Colombia’s independent media, public institutions, and stories were never corrected. And even when and civil-society organizations will face similar chal- media debunked false content, their work did not have lenges to those they encountered in 2018. These a proportional reach when compared to the examples stakeholders should work in collaboration to ramp of disinformation themselves, particularly not in the up efforts to better reach audiences impacted by dis- digital spaces where disinformation and misinforma- information and to more effectively understand the tion circulated. content, formats, and channels necessary for bridging the gaps between where news is published and where Civil-society Organizations and users consume it. Second, Colombia’s Congress is weighing decisions Academic Institutions on several bills that seek to regulate social media, and ome universities and think tanks implemented the Constitutional Court began a public debate on projects aimed at understanding disinforma- the matter before ruling on a libel case pertaining to Stion in Colombia on a deeper level. At the uni- social media. In considering solutions to disinforma- versity level, the production of results, however, is still tion, Colombia faces a dilemma: To address disinfor- a work in progress, and its diffusion has been limited mation, should the country strengthen regulations at to academic circles. Think tanks have similarly taken the risk of infringing on freedom of speech, or should action, but—unlike universities—they have taken steps the country preserve existing laws that do not account to assure the work in seen more broadly, by the pub- for Colombia’s new information environment and that lic and the media. Linterna Verde authored a report are still insufficient for prosecuting cases of illegiti- about disinformation on WhatsApp and began con- mate online activity? The outcome of this debate will versations with Colombian civil society and media be hugely consequential for the Colombian informa- about the matter.166 MOE conducted a study of 44 mil- tion ecosystem—in elections and beyond. lion pieces of content on social media about the elec- Finally, while automation did not play a central role tions. While disinformation was only one of their top- in the 2018 presidential elections, other instances of ics, their findings reached the media. coordinated social media behavior might surface in There were many debates about disinformation the future. From “deep fakes” (i.e., false videos featur- organized by civil-society organizations in Colombia. ing a person doing something they did not do or say- The National Democratic Institute led a series of ing something they did not say) to “cyborg” accounts encounters and conferences with researchers and that feature both bot-like and human behaviors, experts. In their advocacy role, MOE asked candidates researchers and journalists will have to prioritize the to refrain from spreading false information. There were detection and understanding of new instances of such some efforts like No Como Cuento or Convivencias en activity in the future.

41 disinformation in democracies

RECOMMENDATIONS The Atlantic Council’s Adrienne Arsht Latin America Center and DFRLab pro- pose the following recommendations for consideration by government, technology and social media companies, telecommunication companies, fact-checking organizations and the media, civil-society organizations and academic institutions, and international institutions. These suggestions aim at addressing the core, collective challenges posed by disinformation, misinformation, and automation in a rapidly changing information environment.

Actor Challenge Proposed Solution How

Government Electoral authorities Electoral authorities can Electoral authorities can more are at times unable to provide faster responses frequently communicate with verify disinformation, to disinformation and journalists and fact-checkers, misinformation or misinformation and be streamline internal communication automation around more transparent about to allow for faster response time, electoral processes in steps being taken to and publish more details about a timely manner. Some investigate and address steps being taken to debunk claims go completely vulnerabilities. disinformation or misinformation unaddressed or are and fix vulnerabilities during addressed only after the campaign periods. This can be done election has taken place. in partnership with the police, media This can exacerbate and digital influencers. distrust in the voting system.

Law enforcement and Law enforcement and Law enforcement and prosecutors prosecutors are, in some prosecutors can be can dedicate more resources to instances, the only actors more transparent about investigations that involve electoral who can verify claims ongoing investigations disinformation and more clearly that require an official and their respective communicate findings to the public investigation. results, when doing with the help of media organizations so does not affect and local government officials. the legitimacy of said investigations.

Electoral laws in many Government bodies, Government bodies, like Offices of countries prevent like Offices of the the Attorney General in different federal, state, and Attorney General in countries, can better define and local governments different countries, can provide more clarity about which from publishing some establish more clear pieces of information can be information during the guidance about existing published by state institutions electoral period. Media laws that prevent during the electoral period. These black-out laws also public institutions from clarifications can be published online prohibit media from publishing information and divulged to the public ahead of reporting on election- in the days preceding and during elections with the help related issues. Such an election and begin of media organizations and local laws render people considering whether government officials. Media should unable to reference an black-out laws could be allowed to report on election- official source to verify be restructured for a related disinformation up to the day information. digital-first environment of the election. wherein voters increasingly rely on social media and the web for information during elections.

42 recommendations

Actor Challenge Proposed Solution How

Government Electoral laws have not Electoral authorities, At the national and subnational level, caught up to trends police, prosecutors, and governments can debate laws about (continued) in digital information the judiciary should transparency of ad spending online. consumption. examine potential When relevant, governments can Prosecutors have a updates to slander and also partner with technology and difficult time trying libel laws that apply to social media companies to publish cases that involve digital a digitally connected reports holding political parties at forensics or do not have world. all levels of government responsible clearly outlined penalties for their media spending and buying, for actors pushing especially where it pertains to the disinformation or hiring of marketing companies engaging in automation. to send political advertisements through encrypted messaging platforms.

Disinformation and Electoral authorities can Electoral authorities can create misinformation about the form an interdisciplinary multi-stakeholder task-forces to next cycle of local and group to discuss possible discuss disinformation in local municipal elections will measures based on elections and further train staff in soon begin to appear. lessons learned from cities and municipalities to recognize the national elections of and verify false information in 2018. partnership with independent third parties, like civil-society organizations.

Technology and Social Barriers to entry exist Encrypted messaging Encrypted messaging platforms Media Companies on encrypted messaging platforms can provide can consider sharing information platforms for journalists further information that paints a picture of usage on and researchers who about how the platforms the platform–for example, how wish to more effectively work and what actions many groups exist, how many have study and understand are being taken, or could public links and the size of groups disinformation and how be taken, to deter the according to categories–to allow such platforms can serve spread of disinformation. for the independent investigation as vectors for its spread of disinformation by fact-checkers during election cycles. and the media. Platforms can also consider communicating directly with its users about the threat of disinformation, for instance by providing public service announcement push notifications.

Some companies and Technology and social Encrypted messaging platforms, in platforms exhibit both media companies, better understanding the effects of features of one to in understanding the features they offer users, can one messaging and the implications adjust options to minimize broadcast broadcasting. The for the spread of features that increase the potential encrypted nature of disinformation of point for disinformation to go viral with the platforms, along to point communications no pathways for fact-checkers and with broadcast features platforms vs. broadcast journalists to trace their spread. that enable messages platforms, can work to to go viral, can open adjust product features doors for the spread of and functions to account disinformation that is for vulnerabilities of both difficult to track and broadcast messages. difficult to source.

43 disinformation in democracies

Actor Challenge Proposed Solution How

Technology and Social The guidelines that Technology and social Technology and social media Media Companies govern user behavior on media companies companies can streamline and better some technology and can provide clearer communicate clear guidelines about (continued) social media platforms guidelines about the their terms of service and share are not always clear to rules and terms of more information about pages and all users, leading some service that apply to accounts that are taken down with groups and individuals user behavior on their the public to increase understanding to at times claim they platforms. of what constitutes inauthentic are unfairly censored by behavior. The amount of information technology and social to be shared can be weighed to media on the respective account for potential security risks platforms. and risks to privacy.

Technology and social Technology and social Technology and social media media companies media companies companies can capitalize on sometimes lack can establish more channels of communication that immediate and on-the- alliances with local were opened with third parties ground local knowledge institutions and civil- during the 2018 elections to deepen about national or society organizations and expand their knowledge of on- subnational political to better understand the-ground political developments contexts and country- local challenges of and information consumption. specific disinformation disinformation. Companies can open or expand flows. country offices in key markets to speed up the time it takes to address violations of terms of service.

Telecommunications Telecommunications Telecommunications Telecommunication companies can Companies companies’ zero-rating companies, along with work with technology and social policies have created technology and social media companies and fact-checkers incentives for social media companies, can to discuss the possibility of including media users to remain re-examine zero-rating links to fact-checking websites in in a closed online policies through the lens zero-rating policies, so users can space within platforms, of disinformation. have access to more resources when rendering it difficult for engaging with information. them to verify claims using external resources.

Fact-checking Fact-checkers have a Fact-checking agencies Media can experiment with new Agencies and difficult time reaching and independent media framing of content verifying Independent Media audiences most affected can consider new disinformation to learn what is most by disinformation or content and formatting clear and best resonates on social those that rely mostly that better resonates media. They can dedicate more time on encrypted messaging with and better engages to measuring audience sentiment platforms for news. users most pre-disposed and reactions to these experiments to disinformation on to tailor messaging in the most social media. effective way.

Fact-checking Fact-checkers can better Fact-checkers can improve their organizations and the verify narratives rather use of social listening and analytics media at times amplify than only instances tools for a more comprehensive disinformation in their of disinformation, understanding of disinformation attempts to address and better target consumption and groups that might individual cases—this can verified articles to the find value in receiving fact-checked magnify the problem and audiences affected by articles. exacerbate distrust. disinformation.

44 recommendations

Actor Challenge Proposed Solution How

Fact-checking Not all fact-checking Both fact-checkers and Media can ensure that all relevant Agencies and organizations and the media can work to parts of a debate are portraited in Independent Media independent media are increase transparency articles or coverage. Stories should trusted by the public. in reporting and continue to be thoroughly sourced (continued) methodology to bolster and fact-checked. Mistakes should credibility. Independent be corrected and publicized in a media can also continue timely manner. to adhere to the highest standards of journalistic integrity.

Civil-society Academic findings about Academic researchers Academia can more frequently Organizations and disinformation do not can improve their partner with the media to publicize Academic Institutions always reach or inform outreach to media and results of their studies and maintain the public debate. the broader public. direct conversations with civil- society organizations and politicians.

Civil society and Civil society and Civil-society and academia, academia, in analyzing academia can in continued partnership with disinformation, are at continuously target technology and social media times limited in their research and pursue companies, including encrypted ability to target research or expand upon messaging platforms, can work to at the digital spaces partnerships that open new doors for analyzing the where information is enable them to study information environment in the consumed in different user behavior where digital spaces and platforms where locations. information is being users are most actively consuming consumed. information.

International Disinformation is often International International organizations can Organizations transnational in nature organizations can better train election monitors to and can be a threat continue to include understand disinformation flows and to democracies—this efforts to combat dynamics and continue supporting type of threat is not disinformation as part cross-cutting projects to address the constrained by physical of election monitoring issue across regions. borders. initiatives and continue funding local research, advocacy, educational and media projects aimed at combatting disinformation.

45 disinformation in democracies

Conclusion

n Latin America, a region where inequality per- sists and large segments of the population do not yet have enough to eat, digital engagement, especially on messaging platforms, is soaring. Around I215 million people use the Internet at least once per month from six countries in the region: Argentina, Brazil, Colombia, Mexico, Peru and Chile. Of mobile phone Internet users in Brazil, 93 percent actively use WhatsApp. Brazil and Mexico alone comprised 38 percent of WhatsApp users in 2018. This is indicative of a massive shift in information consumption. As the region, and other corners of the world, approach ever-higher rates of connectivity, time is of the essence. Disinformation and misinformation are problems nested in broad contexts—and in trying to minimize the negative impacts of both, actors across the board will also have to assure freedoms of expression and privacy are upheld. In Latin America’s case, where research shows organic disinformation and automation far outweighs Facebook’s vice president for Latin America, Diego Dzodan, any influence by foreign government actors, faster gives an interview for Reuters in and more transparent multi-stakeholder action aimed São Paulo, Brazil in 2017. at addressing false narratives will be key to combatting disinformation and misinformation in an environment of polarization. Likewise, more clarity not only on existing regulations but on attempts to solve the challenges of disinformation can amplify trust and reaffirm democratic legitimacy. The challenges disinformation, misinformation, and automation pose to society do not exist in a silo. These challenges are also not restricted to Latin America. As the world prepares for elections in 2019 and beyond, government, the private sector, civil-society, media and more must work collaboratively to share lessons and best practices. Disinformation, while often uniquely manifested in every country, is a borderless challenge with global repercussions. A collective approach is the only viable solution.

46 REUTERS/Paulo Whitaker Acknowledgments

ust as today’s information environment transcends borders, so too did the efforts to combat disinformation and to strengthen digital resilience as part of #ElectionWatch Latin America. This initia- Jtive included a host of actors, from civil society, to media, to electoral authorities, to the private sector from the United States, Europe and Latin America. For their partnership and support, the Atlantic Council’s Digital Forensic Research Lab (DFRLab) and Adrienne Arsht Latin America Center would like to acknowledge the contri- butions of the many stakeholders and in-country partners that helped make this publication a reality. In Brazil, a special thank you to our partners Fundação Getulio Vargas’ Department of Public Policy Analysis (FGV- DAPP) and Jornal O Globo, specifically the “Fato ou Fake” team. In Colombia, thank you to our partners La Silla Vacía, Linterna Verde, and Centro Gabo. In Mexico, our gratitude goes out to our partners Animal Político and Verificado 2018. Thank you also to the many stakeholders who provided their insight and feedback through consultations: The Organization of American States; in Mexico, the Instituto Nacional Electoral; in Brazil, the Instituto Tecnologia & Equidade, ITS Rio, Internet Lab, the Universidade Federal de Minas Gerais, the Universidade Federal do Rio de Janeiro’s Media Lab, and others; in Colombia, Chicas Poderosas, and many more. The Atlantic Council would also like to recognize Juan Felipe Celia, whose leadership at the inception of this project paved the way for the efforts laid out in this report. Thank you to Paula Garcia Tufró and Nicholas Yap for their efforts and support, as well as to Ben Nimmo, Fernando Meneses, Iain Robertson and the editors that contributed to this publication. A special thank you to Roberta Braga and Maria Fernanda Pérez Argüello for their leadership and manage- ment of this paper. Finally, the Atlantic Council extends a heartfelt thank you to Open Society Foundations (OSF), whose support of the Atlantic Council’s #ElectionWatch Latin America work made possible this comprehensive effort to strengthen digital resilience in Latin America and beyond. The work continues!

47 disinformation in democracies

Glossary of Key Terms

Automated traffic—social-media traffic primarily Disinformation—false information spread with driven by automated accounts, like botnets (see intent to deceive. To prove that a particular inci- definition below). dent counts as disinformation, it is necessary to prove both that the information was false and that Bot—a social media account that has been auto- the source spread it deliberately. mated to carry out certain actions, typically liking, sharing, reposting, or following other users. Bots Election manipulation—an attempt to influ- do not usually engage other users in conversation. ence the outcome of an election through covert, Bots can be further broken down into commercial, undeclared, or illegal means, or by disinformation. political, and spam bots. Election manipulation can be domestic or foreign in origin. It is distinguished from electioneering by Commercial bot—a bot whose activity can be the use of covert, undeclared, or illegal means, and rented out to anyone willing to pay for likes or by the use of disinformation. reposts. Commercial bots are usually apolitical, and often repost or like content on a wide range Fake news—this term has become so debased of unrelated subjects, and in many languages. that it can no longer be assigned a definition.

Political bot—a bot created to amplify one partic- Foreign interference—an attempt to adversely ular political message or stance. Political bots usu- affect, or undermine confidence in, any polit- ally post on one subject, often at a high rate. ical, governmental, or democratic process, or prevent the exercise of human or democratic Botnet—a group of bots organized by the same rights, through coercion, corruption, or the use of person or algorithm, so they behave in the same covert, malicious, or deceptive means, acting from way at the same time. Botnets can consist of hun- abroad. dreds of thousands of accounts, although these are becoming increasingly rare. Inauthentic behavior—behavior on social media that is conducted for a reason other than the Coordinated inauthentic behavior (CIB)—online apparent one. Examples of inauthentic behav- behavior intended to deceive audiences about ior include selling likes, following other accounts its authors or motives, conducted by a number for pay, and setting up anonymous accounts to of users in coordination with one another. CIB is covertly promote other organizations. banned on Facebook, and was the reason used for removing a number of Russian and Iranian disin- Like farm—an organization that uses a substan- formation assets. tial number of social media accounts to sell likes, reposts, and follows to real users who want to Coordinated online network—a group of online make themselves look more important. Like farms users, web pages, social media, or Internet forum are typically commercial, but the return is low: accounts, organized by the same idea, organi- a like farm exposed in 2018 by DFRLab sold ten zation, or person. Members of a network push thousand likes on Instagram. Like farms can use out the same narrative, or narratives; have similar automated accounts, human users, or any combi- behavior patterns; often endorse each other and nation of the two. share the same content, sometimes at different times. Sometimes, but not always, the creation date of accounts is the same.

48 glossary of key terms

Media outlet—not all media outlets subscribe to the Misinformation—false information spread with intent same standards. They can be classified in the following that cannot be ascertained, or which can be shown not ways: to be deliberate (e.g., if the source subsequently corrected itself). Alternative media outlet—a media outlet that contrasts itself to mainstream media and offers its Organic content—content generated by human users. readers an “alternative” take on events and processes. This is one of the most common ways in Organic traffic—social media traffic that appears to be which disinformation actors attempt to win an primarily generated by spontaneous users, rather than audience. a coordinated campaign. (Given the nature of online conversations, even organic traffic is likely to have Commercial media outlet—a media outlet whose some automated component—for example, by adver- primary goal is to earn money for its shareholders. tising bots picking up a popular hashtag to increase It may or may not show independence. users’ awareness of the product being promoted.)

Fringe media outlet—a media outlet that does not Troll—a human being who systematically posts inflam- have the qualities of the mainstream media. This matory, divisive, hyperpartisan, or abusive content, divergence from the mainstream may lie in the size often under cover of anonymity. The word “troll” is not of its distribution and following, or its adherence to a measure of attribution, but describes the behavior of journalistic standards, or it may flow from a self-de- an individual user. clared opposition to the “mainstream media” or “MSM.” Such outlets may or may not share conspir- Troll farm/factory—an organization that employs a acy theories or aggregate and amplify other media substantial number of trolls, usually to promote a polit- outlets—including mainstream media, other fringe ical stance or attack its opponents. Troll operations can media, and blogs. be located in a single website or a number of websites, or organized remotely. Independent media outlet—a media outlet that retains editorial independence, regardless of its source of funding.

Public broadcaster—a media outlet that receives public funding but retains editorial independence from a government or political party.

State-backed media outlet—a media outlet that is partly owned by the government or is owned by a state-owned company.

State-owned media outlet—a media outlet that is overtly or covertly funded by the government.

49 disinformation in democracies About the Authors

Roberta Braga, Associate Director, Adrienne Arsht Latin Jose Luis Peñarredonda, Digital Research Assistant, America Center, Atlantic Council Digital Forensic Research Lab (DFRLab), Atlantic Roberta Braga is an associate director at the Atlantic Council Council’s Adrienne Arsht Latin America Center, where Jose Luis Peñarredonda is a digital researcher and she leads projects on Brazil, disinformation and journalist based in Bogotá, Colombia. His main area of misinformation in elections, trade and commerce, interest is the intersection of technology, politics and and energy in Latin America. In partnership with the culture. He is a digital research assistant at the Atlantic Council’s Digital Forensic Research Lab (DFRLab), Council’s Digital Forensic Research Lab. In the past, he Roberta has for the past year and a half managed the has worked with civil society institutions like Tactical Center’s #ElectionWatch Latin America project focused Tech, Fundación Karisma and the Gabriel García Marquez on exposing disinformation around elections in Brazil, Foundation. His journalism has been published on the Mexico, and Colombia. Braga previously worked as a BBC, NPR, and several Latin American media outlets. He strategic communications analyst at the US Department has a master’s degree in digital culture and society from of Homeland Security, and as a corporate affairs and King’s College London. public relations specialist at Promega Corporation, an international biotechnology firm headquartered in Madison, Wisconsin. She has been quoted in The New Maria Fernanda Pérez Argüello, Associate Director, York Times, The Wall Street Journal, The Financial Times, Adrienne Arsht Latin America Center, Atlantic Council Axios, Brazil’s O Globo and Folha de S.Paulo. Originally Maria Fernanda Pérez Argüello is associate director at the from Brazil, Braga is a native Portuguese- and English- Atlantic Council’s Adrienne Arsht Latin America Center, speaker, and fluent in Spanish. She has a master’s where she leads the Center’s work on Mexico, USMCA, degree in global communication and public diplomacy and Central America, and contributes to projects on from the George Washington University’s Elliott School China- Latin America, regional trade integration as well of International Affairs. as disinformation in Latin America. During her time at the Council, she has co-led the Center’s Central America Task Force, managed the Center’s trade portfolio, and Luiza Bandeira, Digital Research Assistant, Digital programmed events in Asia for US policymakers. She Forensic Research Lab (DFRLab), Atlantic Council speaks native Spanish, English, and French, fluent Italian, Luiza Bandeira is a digital research assistant and the and near-fluent Portuguese. Originally from Costa Rica, lead researcher on Brazil at the Atlantic Council’s Digital Maria Fernanda earned a degree in European Studies Forensic Research Lab (DFRLab), a hub of analysts from Sorbonne University in France, as well as a master’s tracking events in governance, technology, security, and degree in Latin American studies and political economy where each intersect. As part of the #ElectionWatch from Georgetown University’s School of Foreign Service. Latin America project in Brazil, Bandeira identified, She has a master’s in international law and human rights exposed and explained cases of disinformation around from the United Nations University for Peace. the country’s consequential 2018 elections, where she also monitored the spread of disinformation on WhatsApp in conjunction with the Latin American- Donara Barojan, Assistant Director for Research and based group Chicas Poderosas. A multi-media journalist Development, Digital Forensic Research Lab (DFRLab), with over ten years of experience covering elections, Atlantic Council political reforms and economics, Bandeira has written Donara Barojan is an assistant director for research and about foreign affairs and technology for the BBC World development at the Atlantic Council’s Digital Forensic Service and for Brazil’s leading newspaper Folha de Research Lab. She is based at the NATO StratCom S.Paulo. Luiza Bandeira has a master’s degree in social Center of Excellence in Riga, Latvia. At the DFRLab, policy and development from the London School of Donara analyses online disinformation campaigns Economics and has worked as a researcher for the around the world with a special focus on disinformation Harvard Business School’s Latin America program. targeting elections. She also leads efforts to develop tools to monitor and counter disinformation online. Barojan’s research has been featured by numerous media outlets, including The New York Times, Reuters, the Economist, The Hill, ProPublica and Business Insider. Donara is a graduate of the London School of Economics and Political Science.

50 51 disinformation in democracies Endnotes

1 “Number of mobile phone internet users in Brazil from 2015 to 22, 2018, https://politica.estadao.com.br/noticias/ele- 2022 (in millions),” Statistica, 2019, https://www.statista.com/ icoes,entenda-o-que-sao-e-quais-as-diferencas-entre-o- statistics/259749/mobile-phone-internet-users-in-brazil/. fundo-eleitoral-e-o-fundo-partidario,70002362544. 2 “Atlas de Violência,” Instituto de Pesquia Econômi- 18 “Urna eletrônica completa 20 anos de uso no Bra- ca Aplicada, http://www.ipea.gov.br/atlasviolencia/. sil,” Governo do Brasil, August 30, 2016, http://www. 3 Luciana Amaral and Eduardo Militão, “Um capitão no Planal- brasil.gov.br/cidadania-e-justica/2016/07/urna-eletron- to,” UOL, October 28, 2018, https://www.uol/eleicoes/espe- ica-completa-vinte-anos-de-uso-no-brasil. ciais/jair-bolsonaro-eleito-presidente-eleicoes-2018.htm. 19 “Processo eleitoral no Brasil,” Tribunal Superior Eleitoral, 4 BBC News, “Brazil corruption scandals: All you 2018, http://www.tse.jus.br/eleitor/processo-eleitoral-bra- need to know,” April 8, 2018, https://www.bbc. sileiro/funcionamento-do-processo-eleitoral-no-brasil. com/news/world-latin-america-35810578. 20 “Horário eleitoral: Como é feita a divisão do tempo de 5 Luciana Amaral and Eduardo Militão, “Um capitão no Planal- TV,” Huffington Post, last updated on August 23, 2018, to”, UOL, October 28, 2018, https://www.uol/eleicoes/espe- https://www.huffpostbrasil.com/2018/07/20/entenda-como-e-feita- ciais/jair-bolsonaro-eleito-presidente-eleicoes-2018.htm. a-divisao-do-tempo-de-tv-no-horario-eleitoral_a_23485856/. 6 Ibope Inteligência, “Confiança do brasileiro nas instituições 21 Renata Livia Arruda de Bessa Dias, “O Ministério Público é a mais baixa desde 2009”, August 9, 2018, http://www. Eleitoral,” Tribunal Superior Eleitoral, http://www.tse.jus.br/o-tse/ ibopeinteligencia.com/noticias-e-pesquisas/confianca-do-bra- escola-judiciaria-eleitoral/publicacoes/revistas-da-eje/artigos/ sileiro-nas-instituicoes-e-a-mais-baixa-desde-2009/. revista-eletronica-eje-n.-3-ano-3/o-ministerio-publico-eleitoral. 7 FGVDAPP, Disinformation in the Digital Era: Amplifica- 22 Jéferson Botelho, “Cabe à Polícia Civil coibir crimes tions and panorama of the 2018 elections, 2018, https:// eleitorais,” Consultor Jurídico, October 4, 2010, https:// observa2018.com.br/wp-content/uploads/2019/01/ENG-Dis- www.conjur.com.br/2010-out-04/competencia-poli- information-in-the-digital-era-Policy-Paper-2.pdf. cia-civil-coibir-praticas-eleitorais-criminosas. 8 Rodrigo Carro, “Urban Brazil,” Digital News Report, 2018, 23 “Digital Democracy Room #Observa2018,” FGV http://www.digitalnewsreport.org/survey/2018/brazil-2018/. DAPP, https://observa2018.com.br/en/. 9 Mariana Lima, “Brasil já tem mais de um smartphone 24 Márcio Moretto Ribeiro and Pablo Ortellado, “O Que São e ativo por habitante, diz estudo da FGV,” O Estado de Como Lidar Com As Notícias Falsas,” Sur-Revista Interna- S. Paulo, April 19, 2018, https://link.estadao.com.br/no- cional de Direitos Humanos V. 15 N.27 (2018): 71-83, https:// ticias/geral,brasil-ja-tem-mais-de-um-smartphone-ati- sur.conectas.org/wp-content/uploads/2018/07/sur-27-por- vo-por-habitante-diz-estudo-da-fgv,70002275238. tugues-marcio-moretto-ribeiro-pablo-ortellado.pdf. 10 Brazilian Internet Steering Committee, “Survey on the use 25 Page 26, literature review, Fais et al (2017). of information and communication technologies in Bra- 26 Luiza Bandeira, “#ElectionWatch: Fast and False in Brazil,” zilian households,” Núcleo de Informação e Coordenação DFRLab, September 19, 2018, https://medium.com/dfrlab/ do Ponto BR - São Paulo, 2018, https://www.cgi.br/media/ electionwatch-fast-and-false-in-brazil-f020ec0d196c. docs/publicacoes/2/tic_dom_2017_livro_eletronico.pdf. 27 Gustavo Maia and Luciana Amaral, “Bolsonaro critica imprensa 11 Notícias sobre política e eleições nas redes sociais”, October 2, parcial e diz que vai democratizar verbas,” UOL, January 7, 2019, 2018 file:///C:/Users/Luiza/Downloads/Redes%20sociais.pdf https://noticias.uol.com.br/politica/ultimas-noticias/2019/01/07/ 12 G1, “TV é o meio preferido de 63% dos brasileiros para se in- bolsonaro-critica-imprensa-verbas-publicitarias.htm. formar, e internet de 26%, diz pesquisa,” G1 Economia, January 28 Luiza Bandeira, “#ElectionWatch: Protests reinterpret- 24, 2017, https://g1.globo.com/economia/midia-e-marketing/ ed online in Brazil,” DFRLab, October 5, 2018, https:// noticia/tv-e-o-meio-preferido-por-63-dos-brasileiros-pa- medium.com/dfrlab/electionwatch-protests-re- ra-se-informar-e-internet-por-26-diz-pesquisa.ghtml. interpreted-online-in-brazil-128c6f1a9b1a. 13 Endnote: Media Ownership Monitor Brazil, “The 29 Luiza Bandeira, “#ElectionWatch: Partisan Debate and World of Globo”, Intervozes and Reporters With- Electoral Violence in Brazil,” DFRLab, October 26, 2018, out Borders, 2017, https://brazil.mom-rsf.org/en/find- https://medium.com/dfrlab/electionwatch-partisan-de- ings/concentration/, accessed March 18, 2019 bate-and-electoral-violence-in-brazil-3a325dc7589c. 14 “Tiragem impressa dos maiores jornais perde 520 mil ex- 30 Téo Takar and Marcelo Freire, “Vídeo de fraude em urna di- emplars em 3 anos,” Poder360, January 31, 2018, https:// vulgado por Flávio Bolsonaro é falso, diz TRE-MG” , October www.poder360.com.br/midia/tiragem-impressa-dos- 7, 2018, https://noticias.bol.uol.com.br/ultimas-noticias/ele- maiores-jornais-perde-520-mil-exemplares-em-3-anos/. icoes/2018/10/07/tse-esta-apurando-denuncia-de-filho-de-bol- 15 “Estatísticas do eleitorado—Consulta por região/UF/mu- sonaro-sobre-fraude-em-urna.htm, acccessed on March 18, 2019 nicípio,” Tribunal Superior Eleitoral, 2018, http://www.tse.jus. 31 João Filho and Nayara Felizardo, “Joice Hasselmamn foi o br/eleitor/estatisticas-de-eleitorado/consulta-quantitativo. ponto de partido de algumas das maiores mentiras da eleição”, 16 “Partidos politicos registrados no TSE,” Tribunal Su- The Intercept Brasil, https://theintercept.com/2018/10/26/ perior Eleitoral, 2018, http://www.tse.jus.br/parti- joice-hasselmann-mentiras/, accessed on March 18, 2019 dos/partidos-politicos/registrados-no-tse. 32 Luiza Bandeira and Jose Luis Peñaredonda, “#Election- 17 Igor Moraes, “Entenda o que são e quais as diferenças en- Watch: Misinformation Swirls Around Bolsonaro At- tre o fundo eleitoral e o fundo partidário,” Estadão, June tack”, September 14, 2018, https://medium.com/dfrlab/

52 endnotes

electionwatch-misinformation-swirls-around-bolsona- was-up-with-whatsapp-in-brazil-f8476814730f. ro-attack-467701e51ea7, accessed March 18, 2019 50 Wellington Soares, “Conheça o o ‘kit gay’ vetado pelo 33 Luiza Bandeira, “#ElectionWatch: Migration to Gab in Brazil,” governo federal em 2011,” Nova Escola, February 1, 2015. DFRLab, September 24, 2018, https://medium.com/dfrlab/ 51 Luiza Bandeira,” #ElectionWatch: What Was Up electionwatch-migration-to-gab-in-brazil-67a1212c4c76. with WhatsApp in Brazil?” DFRLab, January 14, 2019, 34 Mariana Simões, “Grupos pró-Bolsonaro no WhatsApp https://medium.com/dfrlab/electionwatch-what- orquestram fake news e ataques pessoais na internet, diz was-up-with-whatsapp-in-brazil-f8476814730f. pesquisa,” Publica, October 22, 2018, https://apublica. 52 Wellington Soares, “Conheça o o ‘kit gay’ vetado pelo org/2018/10/grupos-pro-bolsonaro-no-whatsapp-orquestram- governo federal em 2011,” Nova Escola, February 1, fake-news-e-ataques-pessoais-na-internet-diz-pesquisa/. 2015, https://novaescola.org.br/conteudo/84/conhe- 35 “End-to-end encryption,” WhatsApp, https:// ca-o-kit-gay-vetado-pelo-governo-federal-em-2011. faq.whatsapp.com/en/android/28030015. 53 Luiza Bandeira,” #ElectionWatch: What Was Up with 36 Juliana Gragnani, “Como planos de celular com Face- WhatsApp in Brazil?,” DFRLab, January 14, 2019, book e WhatsApp ilimitados podem potencializar propa- https://medium.com/dfrlab/electionwatch-what- gação de notícias falsas,” BBC News Brazil, April 16, 2018, was-up-with-whatsapp-in-brazil-f8476814730f. https://www.bbc.com/portuguese/brasil-43715049. 54 José Marques, “Folha é a maior fake news do Brasil, diz 37 Juliana Gragnani, “Como planos de celular com Face- Bolsonaro a manifestantes”, October 21, 2018, https://www1. book e WhatsApp ilimitados podem potencializar propa- folha.uol.com.br/poder/2018/10/folha-e-a-maior-fake- gação de notícias falsas,” BBC News Brasil, https:// news-do-brasil-diz-bolsonaro-a-,manifestantes.shtml. www.bbc.com/portuguese/brasil-43715049. 55 Luiza Bandeira,” #ElectionWatch: What Was Up 38 Rodolfo Borges, “WhatsApp, uma arma eleitoral sem lei,” with WhatsApp in Brazil?” DFRLab, January 14, 2019, El País, October 22, 2018, https://brasil.elpais.com/bra- https://medium.com/dfrlab/electionwatch-what- sil/2018/10/18/tecnologia/1539899403_489473.html. was-up-with-whatsapp-in-brazil-f8476814730f. 39 Google News Lab and Institute for the Future, “Chat Apps: 56 “Fraude nas urnas e ‘kit gay’ têm mais impacto que outras Frontiers and Challenges for Journalism,” January 2018, https:// notícias falsas,” FGV Dapp, last updated November 1, 2018, newslab.withgoogle.com/assets/docs/chat-app-study-en.pdf. https://observa2018.com.br/posts/fraude-nas-ur- 40 Juliana Gragnani, “Como planos de celular com Face- nas-e-kit-gay-tem-maior-impacto-que-outras-no- book e WhatsApp ilimitados podem potencializar propa- ticias-falsas-em-twitter-facebook-e-youtube/. gação de notícias falsas,” BBC News Brasil, https:// 57 Daniel Bramatti, “WhatsApp teve efeito limitado no 1º turno, www.bbc.com/portuguese/brasil-43715049. diz Ibope,” O Estado de S.Paulo, October 24, 2018, https:// 41 Google News Lab and Institute for the Future, “Chat Apps: politica.estadao.com.br/noticias/eleicoes,whatsapp-teve- Frontiers and Challenges for Journalism,” January 2018, https:// efeito-limitado-no-1-turno-diz-ibope,70002560818. newslab.withgoogle.com/assets/docs/chat-app-study-en.pdf. 58 Versão Portuguesa, “Stolen by Whatsapp! Brazil Poll shows 42 Juliana Gragnani, “Eleições com fake news?: Uma semana Election ‘Drowned’ in Fake News,” AVAAZ, October 31, 2018, dentro de 272 grupos políticos no WhatsApp mostra um Brasil https://avaaz.org/act/media.php?press_id=918. dividido e movido a notícias falsas,” BBC News Brasil, October 5, 2018, https://www.bbc.com/portuguese/brasil-45666742. 59 Luiza Bandeira, “#ElectionWatch: Claims of electronic voting fraud circulate in Brazil,” DFRLab, October 7, 2018. 43 WhatsApp white paper, “Stopping Abuse: How https://medium.com/dfrlab/electionwatch-claims-of-elec- WhatsApp Fights Bulk Messaging and Au- tronic-voting-fraud-circulate-in-brazil-b09cdbe6777a. tomated Behavior,” February 6, 2019 60 Do UOL, “Denúncia de fraude em urnas do DF cit- 44 Gabriel Ferreira and João Pedro Soares, “Como fun- ada por PMs não se comprovou,” UOL, October ciona a máquina de WhatsApp que pode eleger Bol- 10, 2018, https://noticias.uol.com.br/comprova/ulti- sonaro,” Agência Andante in Época, October 24, 2018, mas-noticias/2018/10/10/denuncia-de-fraude-em-ur- https://epoca.globo.com/como-funciona-maqui- nas-do-df-citada-por-pms-nao-se-comprovou.htm. na-de-whatsapp-que-pode-eleger-bolsonaro-23180627. 61 Jonatas Boni, “Homem é detido após gravar vídeo ped- 45 Caio Machado and Marco Konopacki, “Computational Power: indo que eleitores colassem tecla de urna, em RO,” G1, Automated Use of WhatsApp in the Elections,” ITS Rio, October October 28, 2018. https://g1.globo.com/ro/rondonia/ 26, 2018, https://feed.itsrio.org/computational-power-auto- eleicoes/2018/noticia/2018/10/28/homem-e-preso- mated-use-of-whatsapp-in-the-elections-59f62b857033. apos-gravar-video-pedindo-que-eleitores-colassem-te- 46 Fernanda Mena, “Estratégia digital da campanha de Bolson- cla-de-urna-em-ro.ghtml Accessed November 9. aro é uma jabuticaba, diz pesquisador,” Folha de S. Paulo, 62 Amanda Pupo, “Jungmann diz que ameaça a Rosa Weber October 22, 2018. https://www1.folha.uol.com.br/poder/2018/10/ ‘obviamente representa um crime,” O Estado de S. Paulo, estrategia-digital-da-campanha-de-bolsonaro-e-uma-jabu- October 16, 2018, https://politica.estadao.com.br/noticias/ ticaba-diz-pesquisador.shtml. Accessed November 7. eleicoes,ministro-da-seguranca-publica-diz-que-ameaca-a-ro- 47 Luiza Bandeira,” #ElectionWatch: What Was Up with sa-weber-obviamente-representa-um-crime,70002549791. WhatsApp in Brazil?” DFRLab, January 14, 2019, https:// 63 Amanda Lemos, Débora Sögur Hous, and Paulo Pas- medium.com/dfrlab/electionwatch-what-was-up-with- sos, “Sem rastro, WhatsApp pauta eleição de 2018,” Folha whatsapp-in-brazil-f8476814730f, accessed on March 1, 2019 de S.Paulo, https://www1.folha.uol.com.br/poder/2018/10/ 48 Felipe Betim, “ Urna eletrônica: um debate sobre se- sem-rastro-whatsapp-pauta-eleicao-de-2018.shtml. gurança usado por Bolsonaro como ameaça,” El País 64 Tribunal Superior Eleitoral, Resolução Nº 23.551, 18 Brasil, September 22, 2018, https://brasil.elpais.com/ de dezembro de 2017, http://www.tse.jus.br/leg- brasil/2018/09/17/politica/1537215816_674175.html. islacao-tse/res/2017/RES235512017.html 49 Luiza Bandeira,” #ElectionWatch: What Was Up 65 FGV DAPP, “Bots e o Direito Eleitoral Brasileiro,” 2018, with WhatsApp in Brazil?” DFRLab, January 14, 2019, https://observa2018.com.br/wp-content/uploads/2019/01/ https://medium.com/dfrlab/electionwatch-what- Bots_Direito_Eleitoral_elei%C3%A7%C3%B5es_2018.pdf

53 disinformation in democracies

66 Geraldo Doca, “Justiça poderá anular candidaturas base- org/articles/counting-down-first-100-days-amlos-presidency. adas em fake news, diz Fux,” O Globo, August 22, 2018, 82 “Numeralia Proceso Electoral: 2017-2018,” Institu- https://oglobo.globo.com/brasil/justica-podera-anular-can- to Nacional Electoral, July 8, 2018, https://www. didaturas-baseadas-em-fake-news-diz-fux-23001183. ine.mx/wp-content/uploads/2018/08/1Numera- 67 Renan Ramalho, Luiz Felipe Barbiéri, Guilherme Mazui, and lia01072018-SIJE08072018findocx-3.pdf. Alexandro Martello, “Rosa Weber diz que TSE está aprendendo 83 Etellekt, Séptimo Informe de Violencia Política en Méx- a lidar com ‘fake News’,”G1, October 7, 2018, https://g1.globo. ico 2018, July 9, 2018, http://www.etellekt.com/reporte/ com/politica/eleicoes/2018/noticia/2018/10/07/rosa-weber- septimo-informe-de-violencia-politica-en-mexico.html. diz-que-tse-esta-aprendendo-a-lidar-com-fake-news.ghtml. 84 Fiscalía Especializada para Atención de Delitos Elec- 68 FGV DAPP and DIREITO SP, Bots and Brazil’s Elec- torales (FEPADE), http://www.fepade.gob.mx/. toral Legal System, January 15, 2019, https://obser- va2018.com.br/wp-content/uploads/2019/02/EN- 85 USMCA was not signed by the government of En- GWEB-Bots-and-Brazils-Electoral-Law-System.pdf. rique Peña Nieto until November 2018. 69 TSE, “Parceria entre Justiça Eleitoral e agências de checa- 86 Corruption Perceptions Index, Transparency International, Jan- gem de fatos evitou disseminação de notícias falsas no 2º uary 29, 2019, https://www.transparency.org/cpi2018#results. turno das eleições,” October 29, 2018, http://www.tse.jus.br/ 87 Jose I. Rodriguez-Sanchez, Ph.D., Measuring Corruption in imprensa/noticias-tse/2018/Outubro/parceria-entre-justi- Mexico, Rice University’s Baker Institute for Public Policy, ca-eleitoral-e-agencias-de-checagem-de-fatos-evitou-dissem- December 2018, https://www.bakerinstitute.org/media/files/ inacao-de-noticias-falsas-no-segundo-turno-das-eleicoes. files/b190ca73/bi-pub-rodriguez-sanchezcorruption-121118.pdf. 70 Letícia Casado and Reynaldo Turollo Jr, “TSE manda re- 88 Andalusia Knoll Soloff, “Mexico is one of the world’s mover da internet 35 ‘fake news’ que atacam Haddad,” most dangerous places for journalists. People are de- Folha de S.Paulo, October 6, 2018, https://www1.folha. manding action.,” NBC News, May 15, 2018, https://www. uol.com.br/poder/2018/10/tse-manda-remover-da-in- nbcnews.com/news/latino/mexico-one-world-s-most- ternet-35-fake-news-que-atacam-haddad.shtml. dangerous-places-journalists-people-are-n874091. 71 Téo Takar and Marcelo Freire, “Vídeo de fraude em urna 89 Article 19, Democracia Simulada, Nada Que Aplau- divulgado por Bolsonaro é falso, diz TER-MG,” UOL, Oc- dir, March 20, 2018, https://articulo19.org/wp-content/ tober 7, 2018, https://noticias.uol.com.br/politica/ele- uploads/2018/03/INFORME-A19-2017_v04.pdf. icoes/2018/noticias/2018/10/07/tse-esta-apurando-denun- 90 “#NoMásAsesinatos: Exigimos garantías para defender el cia-de-filho-de-bolsonaro-sobre-fraude-em-urna.htm. derecho a defender derechos humano, la libertad de expresión, 72 Luiza Bandeira, “#ElectionWatch: Partisan debate and el ejercicio periodístico y justiciar,” Article 19, February 22, electoral violence in Brazil,” DFRLab, October 26, 2018, 2019, https://articulo19.org/nomasasesinatos-exigimos-garan- https://medium.com/dfrlab/electionwatch-partisan-de- tias-para-defender-el-derecho-a-defender-derechos-huma- bate-and-electoral-violence-in-brazil-3a325dc7589c. no-la-libertad-de-expresion-el-ejercicio-periodistico-y-justicia/. 73 Daniel Funke, “A glitch in Facebook’s fact-checking sys- 91 Forbes Staff, “Elección presidencial dispara el gasto en publici- tem isn’t notifying some users who share false posts,” dad digital,” Forbes, June 22, 2019, https://www.forbes.com.mx/ Poynter, January 25, 2019, https://www.poynter.org/fact- eleccion-presidencial-dispara-el-gasto-en-publicidad-digital/. checking/2019/a-glitch-in-facebooks-fact-checking-sys- 92 Julio César Reveles Aguilar, “Gastos de campaña: Anaya el que tem-isnt-notifying-some-users-who-share-false-posts/. reporta mayor gasto; Meade, el que más eventos ha encabeza- 74 Rob Leathern and Emma Rodgers, “A New Lev- do,” Animal Politico, May 14, 2018, https://www.animalpolitico. el of Transprency for Ads and Pages,” Facebook com/2018/05/gastos-de-campana-anaya-el-que-reporta- Newsroom, June 28, 2018, https://newsroom.fb.com/ mayor-gasto-meade-el-que-mas-eventos-ha-encabezado/. news/2018/06/transparency-for-ads-and-pages/. 93 Forbes Staff, “Elección presidencial dispara el gasto en publici- 75 Bruna Souza Cruz, “Whatsapp vai avisar quando men- dad digital,” Forbes, June 22, 2019, https://www.forbes.com.mx/ sagem for encaminhada, para identificar spam,”UOL, June eleccion-presidencial-dispara-el-gasto-en-publicidad-digital/. 8, 2018, https://noticias.uol.com.br/tecnologia/noticias/ 94 Alejandro Medina, “Las benditas redes sociales que le dieron redacao/2018/06/08/whatsapp-vai-avisar-quando-men- ‘like’ a AMLO,” Forbes, November 29, 2018, https://www.forbes. sagem-for-encaminhada-para-identificar-spam.htm. com.mx/las-benditas-redes-sociales-que-le-dieron-like-a-amlo/. 76 Chris Daniels, “Como o WhatsApp combate a desinfor- 95 “TVs outnumber fridges,” The Economist, April mação no Brasil,” Folha de S. Paolo, October 17, 2018, 28, 2011, https://www.economist.com/blogs/ https://www1.folha.uol.com.br/opiniao/2018/10/como-o- americasview/2011/04/mexico’s_census. whatsapp-combate-a-desinformacao-no-brasil.shtml. 96 María Elena Gutiérrez Rentería, “Digitial News Report: Mex- 77 Critina Tardaguila, “Coalizão inédita de checadores fla- ico,” Reuters Institute and University of Oxford, 2018, http:// gra 50 boatos em 48 horas,” Agência Lupa, October www.digitalnewsreport.org/survey/2018/mexico-2018/. 29, 2018. https://piaui.folha.uol.com.br/lupa/2018/10/29/ coalizao-checadores-50-noticias-falsas-48-horas/ 97 María Elena Gutiérrez Rentería, “Digitial News Report: Mex- ico,” Reuters Institute and University of Oxford, 2018, http:// 78 Aos Fatos, “Conheça a robô checadora de Aos Fatos no www.digitalnewsreport.org/survey/2018/mexico-2018/. Facebook,” October 1, 2018, https://aosfatos.org/noticias/ conheca-robo-checadora-do-aos-fatos-no-facebook/. 98 “Redes sociales y elecciones en México 2018, un nuevo paradigma,” 24 Horas, March 21, 2018, https:// 79 Luiza Bandeira and Jose Luis Peñarredonda, “#ElectionWatch: www.24-horas.mx/2018/03/21/redes-sociales-elec- Misinformation Swirls Around Bolsonaro Attack,” DFRLab, Sep- ciones-en-mexico-2018-nuevo-paradigma/. tember 14, 2018, https://www1.folha.uol.com.br/opiniao/2018/10/ como-o-whatsapp-combate-a-desinformacao-no-brasil.shtml. 99 María Elena Gutiérrez Rentería, “Digitial News Report: Mex- ico,” Reuters Institute and University of Oxford, 2018, http:// 80 “Election Guide: United Mexican States,” USAID, 2019, www.digitalnewsreport.org/survey/2018/mexico-2018/. http://www.electionguide.org/countries/id/140/. 100“Reach of leading social networks in Mexico as of April 81 Carin Zissis, “Countinng Down the First 100 Days of AMLO’s 2018,” Statistica, 2019, https://www.statista.com/statis- Presidency,” ASCOA Online, March 4, 2019, https://www.as-coa. tics/449869/mexico-social-network-penetration/.

54 endnotes

101 María Elena Gutiérrez Rentería, “Digitial News Report: Mex- WhatsApp,” The Mazatlan Post, February 20, 2019. ico,” Reuters Institute and University of Oxford, 2018, http:// 121 Luiza Bandeira, “#ElectionWatch: What Was Up www.digitalnewsreport.org/survey/2018/mexico-2018/. with WhatsApp in Brazil?,” DFRLab, January 14, 2019, 102The Electoral Process and Political Dynamics - Mexico, US https://medium.com/dfrlab/electionwatch-what- Library of Congress, http://countrystudies.us/mexico/91.htm. was-up-with-whatsapp-in-brazil-f8476814730f. 122 “4 datos que caracterizan a los electors que votaron por 103Political Parties in Mexico, Rice University’s Bak- López Obrador,” Politico, July 10, 2018, https://adnpolit- er Institute for Public Policy, https://www.bak- ico.com/presidencia/2018/07/10/4-datos-que-caracteri- erinstitute.org/political-parties-mexico/. zan-a-la-mayoria-de-los-electores-que-votaron-por-amlo. 104 Formerly known as the Instituto Federal Electoral, or IFE. 123 Sergio Rincón, “El millonario negocio detrás de los sitios 105 “Sobre el INE,” Instituto Nacional Elector- de ‘fake news’ en México,” Univision Noticias, April 5, 2017, al, https://www.ine.mx/sobre-el-ine/. https://www.univision.com/noticias/america-latina/el-mil- 106 “The Electoral Tribunal of the Federal Judicial Branch Guaran- lonario-negocio-detras-de-los-sitios-de-fake-news-en-mex- tees the Constitutionality of the Elections,” Tribunal Electoral ico?hootPostID=1b32e034908a9d0b581cce979a64fc9f. del Poder Judicial de la Federación de México, June 2012, 124 Kirk Semple and Marina Franco, “Disinformation, seeded https://www.te.gob.mx/publicaciones/sites/default/files// by bots and trolls, invades Mexico’s election,” The Syd- archivos_libros/The%20Electoral%20Tribunal%20of%20 ney Morning Herald, May 2, 2018, https://www.smh.com.au/ the%20Federal%20Judicial%20Branch%20Guarantees%20 world/north-america/misinformation-seeded-by-bots-and- the%20Constitutionality%20of%20the%20elections.pdf. trolls-invades-mexico-s-election-20180502-p4zctx.html. 107 “Electoral Justice Assessment of the 2018 Mexican 125 “Conoce el convenido de colaboración firmado entre el INE Elections,” International Foundation for Electoral Sys- y Facebook,” Instituto Nacional Electoral, February 13, 2018, tems, June 28, 2018, https://www.ifes.org/news/elec- https://centralelectoral.ine.mx/2018/02/13/conoce-el-con- toral-justice-assessment-2018-mexican-elections. venio-de-colaboracion-firmado-entre-el-ine-y-facebook/. 108 “Qué es Verificado?” Verificado,2018, https://ver- 126 “Conoce el convenido de colaboración firmado entre el INE ificado.mx/que-es-verificado-2018/. y Facebook,” Instituto Nacional Electoral, February 13, 2018, 109 For a taxonomy of bots, see Ben Nimmo, “#BotSpot: https://centralelectoral.ine.mx/2018/02/13/conoce-el-con- Twelve Ways to Spot a Bot,” DFR Lab, August 28, 2017, venio-de-colaboracion-firmado-entre-el-ine-y-facebook/. https://medium.com/dfrlab/botspot-twelve-ways-to- 127 “El INE y Twitter contribuyen para que la ciudadanía ejerza um spot-a-bot-aedc7d9c110c. For a taxonomy of botnets, see voto informado,” Instituto Nacional Electoral, March 27, 2018, Donara Barojan, “#TrollTracker: Bots, Botnets and Trolls,” https://centralelectoral.ine.mx/2018/03/27/el-ine-y-twitter-con- DFRLab, October 9, 2018, https://medium.com/dfrlab/ tribuyen-para-que-la-ciudadania-ejerza-un-voto-informado/. trolltracker-bots-botnets-and-trolls-31d2bdbf4c13. 128 Aldo Nicolai, “¿Se puede sancionar a una compañía por 110 Fernando Peinado, “El mercado global de los ‘me gusta’ difundir noticias falsas?,” Verificado, May 13, 2018,https:// falsos,” El Pais, April 7, 2018, https://elpais.com/internacio- verificado.mx/noticias-falsas-sancion-tribunal/. nal/2018/04/03/actualidad/1522769651_850596.html. 129 “Qué es Verificado 2018?,” Verificado, 2018, https:// 111 Ben Nimmo, Donara Barojan, Jose Luis Peñarredona, and Kan- verificado.mx/que-es-verificado-2018/. ishk Karan, “#ElectionWatch: Trending Beyond Borders in Mex- 130 “Defender la verdad… de la verdad,” Excelsior, February ico,” DFR Lab, June 28, 2016, https://medium.com/dfrlab/elec- 25, 2019, https://www.excelsior.com.mx/opinion/igna- tionwatch-trending-beyond-borders-in-mexico-2a195ecc78f4. cio-anaya/defender-la-verdad-de-la-verdad/1298551. 112 Ben Nimmo and Donara Barojan, “#ElectionWatch: Russian 131 Andrei Gomez-Suarez, “Peace Process Pedagogy: les- Bots in Mexico?,” DFRLab, May 27, 2018, https://medium.com/ sons from the no-vote victory in the Colombian peace dfrlab/electionwatch-russian-bots-in-mexico-46003fcab4. referendum,” Comparative Education, 53.3 (2017), 462–82 113 David Alire Garcia and Noe Torres, “Russia meddling in . Mexican election: White House aide McMaster”, Reu- 132 Pablo Medina Uribe, “In Colombia, a WhatsApp Cam- ters, January 7, 2018, https://www.reuters.com/article/ paign against Posverdad,” The Wilson Quarterly (Wash- us-mexico-russia-usa/russia-meddling-in-mexican-elec- ington DC, 2018) https://www.wilsonquarterly.com/ tion-white-house-aide-mcmaster-idUSKBN1EW0UD. quarterly/the-disinformation-age/in-colombia-a- 114 Donara Barojan, “#ElectionWatch: Down Ballot Bots in whatsapp-campaign-against-posverdad/. Mexico” DFRLab, June 7, 2018, https://medium.com/dfrlab/ 133 See, for instance, the initiatives No Como Cuento, (http://www. electionwatch-down-ballot-bots-in-mexico-e1bee023291d. elavisperomov.org/nocomocuento) a media literacy initiative 115 “#ElectionWatch: Bot Battlefield at #AMLOFest,” Atlantic by journalists and civil society organizations; or Centro Gabo’s Council, June 30, 2018, https://www.atlanticcouncil.org/pub- Convivencias en Red. (https://centrogabo.org/proyectos/ lications/articles/electionwatch-bot-battlefield-at-amlofest. convivencias-en-red), an initiative aimed at understanding 116 Alberto Melchor, “Muertos, tirotecos y robo de urnas deja and tackling disinformation and fostering best practices for la elección en Puebla,” E-Consulta, July 2, 2018, http:// social media use in Colombia (a DFRLab’s local partner). www.e-consulta.com/nota/2018-07-02/seguridad/muer- 134 Juliana Ramírez Prado, “El No ha sido la campaña más barata y tos-tiroteos-y-robo-de-urnas-deja-la-eleccion-en-puebla. más efefctiva de la historia,” Asuntos Legales, October 4, 2016, 117 Andrea Rodriguez, “Ahead of Mexico’s presidential election, https://www.asuntoslegales.com.co/actualidad/el-no-ha-sido- fears and warnings over possible fraud,” Chicago Tribune, June la-campana-mas-barata-y-mas-efectiva-de-la-historia-2427891. 29, 2018, https://www.chicagotribune.com/news/nationworld/ 135 Verjel Torrado and Julián José, “Plebiscito por la paz en Colom- ct-mexico-election-fraud-warning-20180629-story.html. bia, una disputa más allá del contenido de los acuerdos,” Marco: 118 McMaster: US has seen signs of Russian ‘subversion and Revista de Marketing y Comunicación Política, 3 (2017), 57–76 disinformation’ in upcoming Mexican election, Business 136 John Irish, “Colombia’s Santos Says Obstacles to Insider, January 6, 2018, https://www.businessinsider.com/ FARC Peace Deal Surmountable,” Reuters. Decem- mcmaster-us-russia-interference-mexico-election-2018-1. ber 1, 2015, https://www.reuters.com/article/us-co- 119 Donara Barojan, “#ElectionWatch: RT y Sputnik lombia-peace-santos-idUSKBN0TK4DL20151201. Hablan Español,” DFRLab, February 12, 2018, https:// 137 Nicolás Galvis Ramírez, Omar David Baracaldo Orjue- medium.com/dfrlab/electionwatch-rt-y-sput- la, Miguel García Sánchez, and Catalina Barragán Liz- nik-hablan-espa%C3%B1ol-bd641bf7d23e. arazo, Barómetro de Las Américas Colombia, Obesr- 120 “This is How Mexicans Spend Their Day on vatorio de la Democracia, Universidad de los Andes,

55 disinformation in democracies

and USAID, April 2017, https://obsdemocracia.org/up- WhatsApp) Is Dynamic: Disinformation and the spread loads/usaid-barometro-de-las-americas-2016.pdf. of political “chains” in Colombia (Bogota, 2018). 138 Registradoria de Colombia, Elecciones 2018, https://elec- 156 Kanishk Karan and Ben Nimmo, “#ElectionWatch: Bots ciones1.registraduria.gov.co/pre_pres_2018/resultados/2html/ and Biased Polls in Pakistani Elections,” DFRLab, July 25, resultados.html and https://elecciones.registraduria.gov.co:81/ 2018, https://medium.com/dfrlab/electionwatch-bots- elec20180311/resultados/99SE/BXXXX/DSE99999.htm and-biased-polls-in-pakistani-elections-e5782b42d095. 139 MINTIC, “Encuesta de Acceso, Uso y Apropiación 157 Ben Nimmo and Anna Pellegatta, “#ElectionWatch: Italy’s de Las TIC Por Parte de Las Mujeres en Colom- Self-Made Bots,” DFRLab, July 15, 2018, https://medium.com/ bia,” Gobierno de Colombia, https://www.mintic.gov. dfrlab/electionwatch-italys-self-made-bots-200e2e268d0e. co/portal/604/articles-64060_recurso_3.pdf. 158 Donara Barojan, “#BotSpot: Bots Target Malaysian Elec- 140 José Luis Peñarredonda, The Influence Indus- tions,” DFR Lab, April 20, 2018, https://medium.com/dfrlab/ try: Personal Data in Colombia’s 2018 Legisla- botspot-bots-target-malaysian-elections-785a3c25645b. tive and Presidential Elections (Berlin, 2018). 159 Jose Luis Peñarredonda, “#ElectionWatch: Fraud Claims in 141 See YanHaas Mood of the Country, May 2017 (Bogota, Colombia,” DRF Lab, June 7, 2018, https://medium.com/dfr- 2017), http://www.yanhaas.com/yanhaaspoll/YanHaas- lab/electionwatch-fraud-claims-in-colombia-51e141f57745. Poll22May2017_Mercadeo.pdf; Dossiernet, “Quiroga Me- 160 Jose Luis Peñarredonnda, “#ElectionWatch: Bees Cause Buzz in dios Analizó La Influencia De Las Plataformas Digitales Colombia,” DRF Lab, June 13, 2018, https://medium.com/dfrlab/ En Las Elecciones Colombianas,” Dossiernet (Buenos electionwatch-bees-cause-buzz-in-colombia-a7ad06e0be7a. Aires, May 23, 2018), https://www.dossiernet.com.ar/arti- 161 QingQian He, “Issue cross-pressures and time of vot- culo/quiroga-medios-analizo-la-influencia-de-las-plata- ing decision,” Electoral Studies 44 (2016): 362- formas-digitales-en-las-elecciones-colombianas/16004. 373, https://www.sciencedirect.com/science/ar- 142 Registradoria de Colombia, Elecciones 2018, ticle/pii/S0261379415300949?via%3Dihub. https://elecciones1.registraduria.gov.co/pre_ 162 Jose Luis Peñarredonda, “#ElectionWatch: Everyday Mis- cpa_20180826/consultas/html/inicio.html. information in Colombia,” DFR Lab, July 20, 2018, https:// 143 Fabián Hernández, personal conversa- medium.com/dfrlab/electionwatch-misinformation-was-ev- tion with the author (2019). eryday-business-in-colombian-election-7b46ab536d9d. 144 Jose Luis Peñarredonda, “#ElectionWatch: Everyday Mis- 163 George Lakoff and Gil Duran, “Trump has turned information in Colombia,” DFR Lab, July 20, 2018, https:// words into weapons. And he’s winning the linguis- medium.com/dfrlab/electionwatch-misinformation-was-ev- tic war,” The Guardian, June 13, 2018, https://www. eryday-business-in-colombian-election-7b46ab536d9d. theguardian.com/commentisfree/2018/jun/13/how-to-re- 145 Jose Luis Peñarredonda, “#ElectionWatch: Everyday Mis- port-trump-media-manipulation-language. information in Colombia,” DFR Lab, July 20, 2018, https:// 164 Carlos Cortés and José Luis Peñarredonda, Politics (On medium.com/dfrlab/electionwatch-misinformation-was-ev- WhatsApp) Is Dynamic: Disinformation and the spread eryday-business-in-colombian-election-7b46ab536d9d. of political “chains” in Colombia (Bogota, 2018). 146 Carlos Cortés and José Luis Peñarredonda, Politics (On 165 Carlos Cortés, personal communication (2019). WhatsApp) Is Dynamic: Disinformation and the spread 166 See, for example, Peñarredonda, José Luis. “Tenemos Que of political “chains” in Colombia (Bogota, 2018). Hablar de WhatsApp.” Arcadia, October 2018. https://www. 147 Jose Luis Peñarredonnda, “#ElectionWatch: Bees Cause Buzz in revistaarcadia.com/periodismo-cultural---revista-arcadia/arti- Colombia,” DRF Lab, June 13, 2018, https://medium.com/dfrlab/ culo/tenemos-que-hablar-de-whatsapp/71573; Cortés, Carlos. electionwatch-bees-cause-buzz-in-colombia-a7ad06e0be7a. “La Política (En Whatsapp) Es Dinámica.” La Silla Vacía. August 148 Jose Luis Peñarredonda, “#ElectionWatch: Fraud Claims in 9, 2018. https://lasillavacia.com/silla-llena/red-de-la-innova- Colombia,” DRF Lab, June 7, 2018, https://medium.com/dfr- cion/historia/la-politica-en-whatsapp-es-dinamica-67394. lab/electionwatch-fraud-claims-in-colombia-51e141f57745. 149 John Irish, “Colombia’s Santos Says Obstacles to FARC Peace Deal Surmountable,” Reuters, Decem- ber 1, 2015, https://www.reuters.com/article/us-colombia-peace-santos-idUSKBN0TK4DL20151201. 150 Jose Luis Peñarredonda, “#ElectionWatch: Fraud Claims in Colombia,” DRF Lab, June 7, 2018, https://medium.com/dfrlab/electionwatch-fraud-claims-in-colombia-51e141f57745. 151 “Más de 44 millones de conversaciones sobre el proceso electoral colombiano se generaron en las redes sociales,” Misión de Obesrvación Electoral, July 17, 2018, https:// moe.org.co/mas-de-44-millones-de-conversaciones-sobre-proceso-electoral-se-generaron-en-redes/. 152 Jose Luis Peñarredonda, “#ElectionWatch: Partisan Me- dia Deepens Divides in Colombia,” DFR Lab, August 12, 2018, https://medium.com/dfrlab/electionwatch-partisan-media-deepens-divides-in-colombia-896eedbe98fe. 153 Christine Schmidt, “How Getting WhatsApp API Ac- cess Helped This Brazilian Team Project Process 70,000 Fact-Checking Questions,” Nieman Journalism Lab, No- vember 30, 2018, http://www.niemanlab.org/2018/11/ how-getting-whatsapp-api-access-helped-this-brazilian- team-project-process-70000-fact-checking-questions/. 154 “Monitor de Whatsapp,” WhatsApp Monitor, updated Janu- ary 13, 2019, http://www.monitor-de-whatsapp.dcc.ufmg.br/. 155 Carlos Cortés and José Luis Peñarredonda, Politics (On

56 endnotes

57 Atlantic Council Board of DirectorsBoard of Directors

CHAIRMAN Myron Brilliant Andre Kelleners Christopher Smith *John F.W. Rogers *Esther Brimmer Sean Kevelighan James G. Stavridis R. Nicholas Burns Henry A. Kissinger Richard J.A. Steele *Richard R. Burt *C. Jeffrey Knittel Paula Stern EXECUTIVE CHAIRMAN EMERITUS Michael Calvey Franklin D. Kramer Robert J. Stevens *James L. Jones James E. Cartwright Laura Lane Mary Streett John E. Chapoton Richard L. Lawson Ellen O. Tauscher Ahmed Charai Jan M. Lodal Nathan D. Tibbits CHAIRMAN EMERITUS Melanie Chen Douglas Lute Frances M. Townsend Brent Scowcroft Michael Chertoff Jane Holl Lute Clyde C. Tuggle *George Chopivsky William J. Lynn Melanne Verveer PRESIDENT AND CEO Wesley K. Clark Wendy W. Makins Charles F. Wald *Frederick Kempe *Helima Croft Zaza Mamulaishvili Michael F. Walsh Ralph D. Crosby, Jr. Mian M. Mansha Geir Westgaard Nelson W. Cunningham Chris Marlin Maciej Witucki EXECUTIVE VICE CHAIRS Ivo H. Daalder Gerardo Mato Neal S. Wolin *Adrienne Arsht *Ankit N. Desai Timothy McBride Mary C. Yates *Stephen J. Hadley *Paula J. Dobriansky John M. McHugh Dov S. Zakheim Thomas J. Egan, Jr. H.R. McMaster VICE CHAIRS *Stuart E. Eizenstat Eric D.K. Melby HONORARY DIRECTORS *Robert J. Abernethy Thomas R. Eldridge Franklin C. Miller James A. Baker, III *Richard W. Edelman *Alan H. Fleischmann *Judith A. Miller Harold Brown *C. Boyden Gray Jendayi E. Frazer Susan Molinari Ashton B. Carter *Alexander V. Mirtchev Ronald M. Freeman Michael J. Morell Robert M. Gates *Virginia A. Mulberger Courtney Geduldig Richard Morningstar Michael G. Mullen *W. DeVier Pierson Robert S. Gelbard Edward J. Newberry Leon E. Panetta *John J. Studzinski Gianni Di Giovanni Thomas R. Nides William J. Perry Thomas H. Glocer Franco Nuschese Colin L. Powell Murathan Günal Joseph S. Nye TREASURER Condoleezza Rice John B. Goodman Hilda Ochoa-Brillembourg *George Lund George P. Shultz *Sherri W. Goodman Ahmet M. Oren Horst Teltschik *Amir A. Handjani Sally A. Painter John W. Warner SECRETARY Katie Harbath *Ana I. Palacio William H. Webster *Walter B. Slocombe John D. Harris, II Carlos Pascual Frank Haun Alan Pellegrini *Executive Committee Members DIRECTORS Michael V. Hayden David H. Petraeus List as of January 1, 2019 Stéphane Abrial Brian C. McK. Henderson Thomas R. Pickering Odeh Aburdene Annette Heuser Daniel B. Poneman *Peter Ackerman Amos Hochstein Dina H. Powell Timothy D. Adams *Karl V. Hopkins Robert Rangel Bertrand-Marc Allen Robert D. Hormats Thomas J. Ridge *Michael Andersson *Mary L. Howell Michael J. Rogers David D. Aufhauser Ian Ihnatowycz Charles O. Rossotti Matthew C. Bernstein Wolfgang F. Ischinger Robert O. Rowland *Rafic A. Bizri Deborah Lee James Harry Sachinis Dennis C. Blair Reuben Jeffery, III Rajiv Shah Thomas L. Blair Joia M. Johnson Stephen Shapiro Philip M. Breedlove Stephen R. Kappes Wendy Sherman Reuben E. Brigety II *Maria Pica Karp Kris Singh

The Atlantic Council is a nonpartisan organization that promotes constructive US leadership and engagement in international affairs based on the central role of the Atlantic community in meeting today’s global challenges.

1030 15th Street, NW, 12th Floor, Washington, DC 20005

(202) 778-4952 www.AtlanticCouncil.org