Meta’s AI content moderation errors have come under renewed scrutiny after election-related incidents in Brazil and the United States, where the company blamed automated systems for restricting political accounts and content. The failures affected supporters of Brazilian president Luiz Inácio Lula da Silva, the US protest coalition No Kings and a Florida Senate candidate’s Latino outreach campaign.
The incidents have emerged as Facebook and Instagram’s parent company accelerates a shift from human reviewers to large language models. Meta announced in March that it would introduce more advanced automated enforcement systems, and by June the technology was handling about half of review requests.
The company is aiming for at least 90% automation in some categories by the end of 2026. Meta has said its models make 13% fewer errors than human reviewers and identify 10% more violations, although some employees have warned that the transition is moving too quickly and lacks sufficient oversight.
An October report by the Institute for Strategic Dialogue said platforms were increasingly relying on crowdsourced and automated systems rather than human oversight. It found that companies had disclosed little about how the systems worked or were tested, describing their use as the primary safeguard during a fast-moving election as “an untested proposition”.
Brazil election dispute puts Meta before court
In Brazil, Lula’s campaign went to the Superior Electoral Court on October 10, alleging that Meta’s algorithms and automated restrictions had limited his online reach while favouring his opponent, Flávio Bolsonaro.
The campaign cited examples including an automatic translation error in which the comment “Flávio futuro presidiário!!” — meaning “Flávio, future prisoner!!” — was rendered as “Flávio future president!!”. It also alleged that some of Lula’s posts had been restricted from being shared.
A Brazilian newspaper later reported that Meta staff had acknowledged in internal messages that Lula’s accounts were more restricted than Bolsonaro’s. Meta said the problems were technical errors and denied treating the candidates unequally.
On October 15, the electoral court ordered Meta to apply its rules equally to both candidates, threatening a daily fine of 500,000 reais, or about 98,000 dollars. After judging the company’s first response inadequate, the court gave Meta 24 hours on October 19 to demonstrate that neither campaign had been restricted.
Meta subsequently listed 17 measures, including human review within two hours of any automated restriction on advertising accounts belonging to the presidential campaigns. Bolsonaro’s campaign later joined the case, saying it too had been treated unfairly.
Brazil’s electoral prosecutor’s office told the court that Meta had not fully complied with the orders. It said the company had failed to explain how its automated tool treated content about each candidate and argued that the errors varied by candidate, weakening Meta’s explanation that they were simply technical faults.
Lula won the October 25 run-off with 51.8% of valid votes. The court has yet to decide whether to fine Meta, while federal prosecutors have opened an inquiry into account removals and filed a civil claim against the company.
No Kings account suspended during US day of action
In the United States, Meta suspended the Instagram account of No Kings on October 17, during the coalition’s fourth national day of action encouraging early voting. Organisers said the account disappeared at about 1.30pm Eastern Time, when events were at their busiest, and remained inaccessible for roughly 90 minutes.
Administrators were unable to access the account for several more hours. The coalition also said that users had reported No Kings and early-voting content as inaccessible, restricted or apparently suppressed across Meta’s platforms.
Meta’s explanation changed the following day. Andy Stone, the company’s vice-president of communications, initially attributed the suspension to unusually high levels of posting, sharing and commenting.
He later said that Meta’s network security systems had detected “the use of a Claude tool reading profiles, posts, and hashtags” in intense bursts. The company’s “automated protections”, he said, had mistakenly taken the account offline.
“We want to reiterate our sincere apology for the error,” Mr Stone wrote. No Kings later confirmed that its social media team had used Claude, an AI model, to search keywords and hashtags and monitor online discussion.
The coalition wrote to Meta chief executive Mark Zuckerberg asking whether the company had communicated with the Trump administration about the event or the 2026 mid-term elections. It also sought details of Meta’s “whitelist protocols”, which protect some high-profile accounts from automated removal, and asked how they treated Donald Trump and his allies compared with No Kings and other groups encouraging voter participation.
Meta apologised again and offered a meeting with technical specialists, according to the coalition. Organisers said the response did not answer their questions and called instead for a meeting with senior leadership.
Five Democratic senators, led by Chris Van Hollen of Maryland, later demanded records of any communications between Meta and the Trump administration about No Kings. They described the company’s changing explanations as alarming.
The coalition said it met senior Meta executives on October 30, but that the discussion ended without substantive commitments to change the company’s practices or introduce safeguards.
Florida candidate’s Spanish-language account removed
Another automated enforcement error affected Angie Nixon, a Democratic candidate for the US Senate in Florida. Instagram removed a Spanish-language account created by her campaign to reach Latino voters as the campaign was posting from a get-out-the-vote rally in Osceola County on October 27.
Ms Nixon called for the page to be restored and for an explanation of its removal. It was back online the next morning.
A Meta spokesperson said the account had been created only about a week and a half earlier and that the company’s systems had mistakenly judged it to be impersonating the Nixon for Senate campaign.
The organising movement 50501 linked the suspension to the No Kings incident, calling for “transparency and accountability beyond technical explanations”.
Pressure grows for answers over election moderation
Meta’s Oversight Board, an independent body funded by the company that reviews content moderation decisions, criticised what it called “recurring enforcement failures in Meta’s election integrity efforts”. It cited the No Kings suspension and the excessive enforcement of fraud rules against Lula campaign accounts.
The board did not suggest that either incident was politically motivated, and Meta is not bound by its statement. Attorneys general from 13 states and the District of Columbia have nevertheless asked the company to explain the No Kings suspension and set out how it distinguishes legitimate political activity from harmful bot traffic.
Robert Garcia of California and Jamie Raskin of Maryland, the senior Democrats on the House Oversight and Judiciary committees, have requested a briefing by November 9.
“We need records explaining how this happened and whether anyone in the Trump Administration had a hand in it,” Mr Raskin said.
The demands do not themselves carry the force of law. In Brazil, the electoral court’s decision on a possible fine and the prosecutors’ civil case will test whether the government can compel Meta to provide fuller answers.
In the United States, the extent of congressional scrutiny could depend on the vote. If Democrats win the House, Mr Garcia and Mr Raskin would be in line to chair committees with the power to subpoena answers to some of the same questions.
