Meta Admits Lula Glitches, Denies Bias

Meta admitted three temporary technical problems connected to Lula’s campaign. On 23 September, an automated check wrongly identified an advertisement as impersonating a government social programme. That mistake switched Lula’s advertising account off for just over one hour.

On 7 October, a bug froze video-view counts in more than 100 countries for about five hours. Meta said Flávio Bolsonaro’s campaign reported the same fault that day. Around 9 October, a code update briefly stopped thousands of advertisers in several countries from creating ads from existing posts, although direct uploads still worked.

Meta described the incidents as technical errors that affected users or advertisers across countries, rather than actions aimed at Lula. Lula’s coalition nevertheless argues that the faults reduced his campaign’s reach. The electoral dispute remains important because the court ordered Meta to treat both candidates equally in distribution, recommendations and search.

Brazil’s Superior Electoral Court, or TSE, is the country’s court for electoral disputes and rules. In this case, Lula’s coalition brought a lawsuit claiming that Meta reduced the reach of his posts and shut his advertising account. The court therefore became the authority deciding whether the platform was following Brazil’s election rules.

The TSE can order equal treatment because the dispute concerns political communication during an election. On 15 October, Justice Marco Aurélio Bellizze partly granted the coalition’s request. He ordered Meta to treat both candidates equally in distribution, recommendations and search. The order included a fine of 500,000 reais, about US$99,000, per day.

The ruling does not prove that Meta deliberately manipulated content. The article says an earlier decision found no proof of manipulation at that stage. It shows instead that Brazil’s electoral court can set requirements for a platform’s political reach while the case continues.

The admitted technical problems were limited in duration but potentially broad in reach. Lula’s advertising account was disabled for just over one hour on 23 September. A video-view-count bug then lasted about five hours on 7 October and affected more than 100 countries. Around 9 October, a code update affected thousands of advertisers in several countries.

The engagement gap was also large. A survey by Lula’s Workers’ Party communication secretariat found that Flávio Bolsonaro’s page received 2.96 times more engagement on average. It reportedly did so despite having 17.2% fewer followers than Lula’s page. Engagement can include actions such as viewing or interacting with campaign content, although the article does not define the survey’s exact measure.

These figures do not by themselves prove political interference. Meta says the technical faults affected other users or campaigns, including Bolsonaro’s campaign. The court dispute focuses on whether the overall pattern nevertheless produced unequal political reach during the runoff campaign.

The immediate consequence is a court-backed requirement, not a finding that Meta intentionally interfered. On 15 October, the TSE ordered Meta to treat Lula and Flávio Bolsonaro equally in content distribution, recommendations and search. The order carries a daily fine of 500,000 reais, about US$99,000.

If Meta failed to follow that order, the stated mechanism would allow the fine to accumulate for each day of noncompliance. The court could use that financial pressure to require equal treatment while the electoral dispute proceeds. The article does not identify any additional penalty or say that the admitted faults themselves already violated campaign rules.

Meta maintains that the incidents were technical errors affecting users in several countries. It also cites an earlier decision saying there was no proof, at that stage, that it manipulated distribution for or against a candidate. The next practical issue is whether Meta can demonstrate compliance with the court’s equality requirements.

A technical error can resemble favoritism when it affects one candidate’s account at a politically important moment. Turning off an advertising account interrupts paid distribution. Freezing view counts can make a campaign’s performance look weaker or delay decisions based on those figures. Even a short interruption may matter when campaigns are competing for attention during a runoff.

The article gives concrete examples. Lula’s account was disabled for just over one hour after an automated system misclassified an advertisement. A separate bug froze video-view counts for about five hours in more than 100 countries. Meta said Flávio Bolsonaro’s campaign reported that same fault, but Lula’s coalition alleges a broader reduction in its reach.

The dispute therefore concerns both actual access and perceived performance. Meta says the errors were widespread, not targeted. The TSE still ordered equal treatment in distribution, recommendations and search, because unequal effects can become politically significant even without proven intent.

The clearest test would be evidence of intent and selectivity. Investigators could examine internal instructions, approval records, code changes, and staff messages to see whether anyone ordered a restriction against Lula or favoured Bolsonaro. They could also compare the timing and impact across candidates, countries, accounts, and content types.

An ordinary bug would more likely match a documented software change or automated rule, affect multiple users, and stop after a fix. Meta points to that pattern: the account decision followed an automated classification error, while the view-count problem affected more than 100 countries. It also says Bolsonaro’s campaign reported the same view-count fault.

A deliberate intervention would be more persuasive if records showed a candidate-specific rule, selective rollback, repeated unexplained effects, or concealment after engineers identified the problem. The article does not report such evidence. It quotes an earlier court decision saying manipulation had not been shown at that stage, so the dispute remains unresolved.

Advertising systems decide which eligible users may see a paid political message. They commonly use campaign settings, audience characteristics, budget, timing, and an automated auction or delivery system. Moderation and policy checks can reject, limit, or pause an ad before it reaches people. The article’s example was an automated check that wrongly classified a Lula ad.

Recommendation systems rank unpaid posts or videos for users, while search systems order results when people look for topics or candidates. Signals can include relevance, predicted interaction, freshness, and policy restrictions. These systems determine visibility rather than simply displaying every post equally. The article says the TSE order covered distribution, recommendations, and search.

Because these systems operate at scale, a coding error or ranking change can affect reach quickly. A view-count bug may change measurement without changing delivery, while an account shutdown directly stops advertising. The article does not describe Meta’s specific algorithms, so this is a general explanation of how such systems typically work.