Facebook Owner Meta Is Failing to Prevent Repeat of Jan. 6 in Brazil, Report Warns

Meta is failing to stop a Jan. 6-style movement gaining traction on Facebook and WhatsApp in Brazil, a rights group has warned in a new report.

Campaigning for Brazil’s next presidential election is currently underway, with the first round of voting scheduled for October 2. Jair Bolsonaro, the country’s rightwing populist president, has embraced comparisons to Donald Trump throughout his tenure and is now drawing from the former U.S. President’s 2020 playbook as he lags behind his leftwing opponent Luiz Inácio Lula da Silva.

In recent months Bolsonaro has sought to sow doubt about the validity of Brazil’s democratic processes, warning repeatedly of the risk of fraud, and alleging without evidence that the country’s electronic voting system is vulnerable to interference. In speeches he has said that “only God” can remove him from office. Some of Brazil’s key military officials have echoed his claims of possible fraud, sparking fears that the world’s seventh most populous country could be vulnerable to a military coup if Bolsonaro is defeated at the polls in October.

“If need be,” Bolsonaro said at an election rally in June, “we will go to war.”

Content questioning the validity of the upcoming election is spreading rapidly on social media, according to a new report titled “Stop the Steal 2.0” that was shared exclusively with TIME by the non-profit watchdog SumofUs. The spread of this content is contributing to a Jan. 6-style movement gaining traction in Brazil—and Meta, which also owns Brazil’s most widely-used messaging platform WhatsApp, is not doing enough to prevent it, according to SumofUs. The findings echo those of another report published by Global Witness in mid August, which found that Facebook repeatedly approved ads containing falsehoods about the Brazilian election.

Social media’s role in Brazil’s elections

The impact of social media on Brazilian politics is hard to overstate. Some 83% of Brazilians get their news online, including via social media, according to the Reuters Institute. WhatsApp is the most popular platform in Brazil, with 78% of people using it regularly; Facebook is used by 67% of the population. “Bolsonaro is the first Brazilian president, probably one of the few leaders in the world, who governs by social media, much more than Trump ever did,” says Thomas Traumann, a Brazilian journalist and political analyst who served as a spokesperson for Dilma Rousseff, one of Bolsonaro’s predecessors as president.

In the report shared with TIME, SumofUs identified posts, ads and private messages on Facebook and WhatsApp that it said contributed to “inciting a violent coup.” The report also identified ads from Bolsonaro supporters spreading electoral disinformation, with targets including Brazil’s Supreme Court and Bolsonaro’s political opponents. SumofUs claims that other ads posted on Facebook in August broke Brazilian laws about political advertising outside of the official campaign period.

Many of the posts called on Brazilians to take to the streets on Sept. 7, Brazil’s independence day, in a planned show of force that the SumofUs report warns could become “a repeat of January 6.”

“Meta has learned absolutely nothing since January 6 in the U.S.,” says Flora Rebello Arduini, a campaign manager at SumofUs and the author of the report. “We are seeing ads that are pushing not just for a violent coup in the country, but also narratives discrediting the electoral processes in Brazil.”

SumofUs identified 16 “problematic” Facebook ads that promoted the Sept. 7 rally, including one that included a picture of a combat knife along with military gloves and goggles. Together, the 16 ads were viewed by Facebook users more than 615,000 times, the report says. (Meta removed the ad with the combat knife before publication of the report, but an identical post on a smaller page is still online, according to SumofUs.) Rebello Arduini says while the report’s sample size is small, the ads it details are “just the tip of the iceberg” of the narrative being developed across social media.

Rebello Arduini adds that Facebook appears to be blind to the wider significance of the Sept. 7 movement, addressing posts on a case-by-case basis rather than as part of a coordinated threat to Brazilian democracy. “You cannot assess specific isolated pieces of content or ads without actually putting them in the context of the country,” she tells TIME. “One size doesn’t fit all, unfortunately.”

Bolsonaro’s supporters are also using WhatsApp to undermine confidence in the election, according to the report. SumofUs monitored three large group chats on the platform over a one-week period and found memes inciting violence on Sept. 7. “A war is not won in hours,” one example reads. “Sept. 7 is just the beginning […] Want freedom? FIGHT. Want your job? FIGHT. Want to protect your family? FIGHT.”

“We cannot comment on a report that we have not been given access to,” a Meta spokesperson said in a statement. “We’ve prepared extensively for the 2022 election in Brazil, working closely with local electoral and law enforcement officials. We remove content and accounts that constitute a credible threat to public or personal safety.” In mid-August, Meta announced it was banning ads that question the legitimacy of the upcoming election, and announced plans to activate an “election operations center” closer to voting day to “identify potential threats in real time and speed up our responses.”

“The concern of violence on the street is absolutely real,” says Katie Harbath, Facebook’s former head of elections who spent time in Brazil for the company during the 2018 election campaign in which Bolsonaro swept to power. Harbath says Facebook has, however, noticeably improved its policies since the last election, by rolling out labels that point users to reliable information from the electoral board, and improving its AI systems for detecting harmful language.

“What has not changed is the difficult question about where you draw the line with some of this content, and the question of when something rises to the level of imminent harm,” says Harbath, who now leads a tech policy consultancy called Anchor Change. “To me, that’s a story of how AI is not nuanced enough yet to know the difference between when somebody says, ‘let’s take to the streets and defend our democracy,’ whether they are intending to be violent or not.”

Meta’s own policies may be making it harder for researchers to point out the company’s flaws. Harbath cautioned against taking the small sample sizes in the SumofUs report as proof that its findings are just the tip of an iceberg. But the report’s authors told TIME that Meta had made it harder for them to do more wide-ranging research by rejecting their requests to access CrowdTangle, a tool that allows researchers to monitor the reach of posts and hashtags. Meta is reportedly planning to shutter CrowdTangle, following a slew of bad press stemming from researchers’ and journalists’ use of the tool. “That limits the ability of researchers to better look through what’s happening on the platform,” Rebello Arduini, the author of the SumofUs report, told TIME. “Facebook is again closing the circle and limiting the possibilities for researchers, civil society, and academics.”

As tensions rise on social media, Brazilian political analysts now worry that even if Sept. 7 remains peaceful, the election could end very badly. “I don’t expect a coup attempt on Sept. 7,” says political analyst Traumann. “Bolsonaro wants to show the people are behind him, and then we’ll see what happens on Oct. 2. The danger will come on election day, the days before, and the days after.”