Meta Accused of Ignoring Brazilian Court Order on Bolsonaro Favoritism: TSE Filing, Fines, and the Algorithmic Transparency Fight

Meta is facing a new filing at Brazil's Superior Electoral Court (TSE) accusing it of continuing to give Sen. Flávio Bolsonaro more favorable treatment than President Luiz Inácio Lula da Silva, despite a court order requiring equal treatment of the two campaigns. The evidence submitted by Sleeping Giants Brasil and CTRL+Z consists of animated GIF search results, hashtag suggestion ordering, and two Meta AI answers that returned the numbers 13 and 22 — the party ballot numbers of Lula and Flávio Bolsonaro — in a test that made no reference to the election.

The filing was made on Monday and reported by journalist Mônica Bergamo in Folha de S.Paulo. The story was originally published in Portuguese by Brasil 247 and translated by Panamerican Dispatch, with a byline for Otávio Rosso.

The petition asks for three things: a 24-hour fix, algorithmic disclosure, and enforcement of the fine

The petition is addressed to TSE Justice Marco Aurélio Bellizze, who is overseeing the case. The organizations ask him to:

• Order the alleged irregularities corrected within 24 hours.
• Require Meta to explain how it ranks search results and how it produces answers with artificial intelligence.
• Enforce the daily fine of R$500,000 (about $99,000) that the court had already set.

The groups are not claiming they have proven bias inside Meta's systems. They say the examples reinforce their suspicion that the platforms do not apply equivalent criteria to the two presidential campaigns, and they are putting the question to the electoral court rather than asserting a conclusion.

The evidence: GIF searches for "brasil" surfaced Flávio Bolsonaro

The petition says a GIF search on WhatsApp for the word "brasil" returned as its first result an image of Flávio Bolsonaro showing his ticket number and identifying his running mate. According to the records sent to the court, the same result appeared on three separate accounts. The groups use this to ask whether the top position comes from systematic ranking criteria.

A similar check on Instagram showed images of the senator among the first results. Users who searched for "brasil" found three images of Flávio Bolsonaro in the top positions. The only mention of President Lula in those results was his name in writing, with no photo.

The organizations argue the difference deserves an explanation because search and recommendation tools can increase the exposure of specific candidates on social media.

The evidence: Lula linked to derogatory terms and prison imagery

A second set of examples ties Lula to negative terms in searches on Meta platforms.

• On WhatsApp, a GIF search for the word "ladrão" (thief) returned a caricature of the president in a striped uniform, a reference to prison dress. Another result carried the message "Supporting the Workers' Party is a mental illness."
• On Instagram, a search for the same word showed a GIF reading "Vote 13," the Workers' Party election number tied to Lula's candidacy.

Sleeping Giants Brasil and CTRL+Z describe these results as an imbalance in how political references are shown to users, and say the imbalance stands out against the visibility of content favorable to Flávio Bolsonaro.

The evidence: hashtag suggestions ranked critical terms differently for each candidate

The petition also challenges how Instagram orders hashtag suggestions. According to the filing, a search for "lula" offered "#lulaladrão" (Lula thief) as the first suggestion, ahead of other terms about the president. A search for "bolsonaro" placed the critical hashtag "#bolsonaronuncamais" (Bolsonaro never again) among its last suggestions.

The groups cite the difference in position as another reason to investigate the relevance criteria the platform uses. Their framing is that the issue is not only whether content for or against the candidates exists, but how automated systems organize that information and present it to the public.

The evidence: two Meta AI answers returned 13 and 22

The organizations included a test of Meta AI, the artificial intelligence tool offered to WhatsApp users. According to the petition, two people asked the system to pick a number between 1 and 25. They gave no instructions and made no reference to the election. In one attempt, the AI answered 13; in the other, it answered 22. Those are the party ballot numbers of Lula and Flávio Bolsonaro, respectively.

The groups question the coincidence and ask for an explanation of the parameters the system uses to form its answers. The article states plainly that the episode alone does not allow a conclusion that the tool answered out of electoral preference, but the organizations list the test among the elements that, in their view, demand more transparency about how the company's AI features work.

The court has already ordered equal treatment and warned the fine could double

Justice Bellizze had already ordered Meta to apply identical criteria to the campaigns of Lula and Flávio Bolsonaro, and to prove it was following the measures. On Friday, the full court upheld the justice's decision, reinforcing the company's duty to ensure equal treatment of the candidates. On Sunday, Bellizze again demanded action, giving the company a new 24-hour deadline to show it had met part of the court's requirements.

The justice also warned that the daily fine, first set at R$500,000 (about $99,000), could rise to R$1 million (about $198,000) if noncompliance continues. The new petition from Sleeping Giants Brasil and CTRL+Z was filed to add to the case just as the court demands more detailed explanations from the company.

The dispute started with complaints from Lula's campaign

The case began with complaints from Lula's campaign about how Meta's digital platforms work. The president's representatives pointed to limits on the circulation of content about his candidacy, and cited differences in how the company's tools behave compared with tools tied to his opponent. Those claims led the electoral court to demand measures ensuring equal treatment of the candidates.

Meta has denied deliberately favoring any candidate and blamed the earlier problems on technical failures. Folha de S.Paulo contacted the company about the new records filed Monday; Meta had not responded by the time of publication. The lack of a specific response leaves the explanations for the mechanisms behind the results unresolved.

The strongest criticism of the filing targets the Meta AI number test

The most repeated objection in the Hacker News discussion is that the numerological Meta AI test weakens the rest of the case. One commenter wrote:

According to the petition, two people asked [Meta AI] to pick a number between 1 and 25. They gave no instructions and made no reference to the election. In one attempt, the AI answered 13. In the other, it answered 22. Those are the party ballot numbers of Lula and Flávio Bolsonaro, respectively. The groups question the coincidence. My confidence in the validity of their claims tanked right about here. I'm all for regulating transparency into social media algorithms. Meta most certainly is compromised by foreign actors. But can we choose less stupid hills to die on than numerology and GIFs? Headline is not remotely supported by the evidence.

A second commenter made the same point about the GIF results, arguing the "brasil" result has an obvious innocuous explanation:

I know nothing about Brazilian politics, but this looks like a desperate attempt to undermine the opposition or pull Meta into needless controversy... So what if searching for "brasil" returns a Bolsonaro GIF? I tried it myself, and the obvious explanation is that he's wearing a Brazilian flag tshirt and dancing rather comically. How does a GIF appearing in search results imply political bias?

The alternative technical explanation: engagement-driven ranking, not political intent

Several commenters argued that the results described in the petition can be produced by ordinary ranking systems rather than deliberate political favoritism. The reasoning is that right-wing content tends to be more emotional and aggressive, which drives engagement, so a standard ranking algorithm will amplify it without any electoral preference being coded in.

I have a feeling the problem isn't that malicious unfortunately. Right-wing folks are driven by social media since ~2016. Their content is emotional, aggressive and naturally conducive to more engagement. It's what their base feeds on. So a lot of the situations described here are just standard ranking algorithms doing their job. But that begs the question: is the existence of these platforms that amplify divisive and inflammatory content the problem itself? I vote yes.

A related argument asks what a platform could even do if the imbalance comes from user behavior rather than platform design. One commenter pointed out that a firehose of newest comments, or a most-upvoted ordering, would still benefit the candidate whose online support outnumbers the opposition, and suggested instead that official accounts of candidates should be immune from suspension except by the highest level of administrators, given that a trillion-dollar company has the resources to monitor high-level official accounts.

A self-reinforcing personalization problem

The petition's screenshots raise a measurement question that one commenter highlighted: if search results and AI invocation are personalized, then people checking whether a search is biased may themselves be influencing the result. According to that comment, Meta has a long history of basing personalization on what people close to a user do — for example, inferring proximity from shared location or WiFi and then letting one person's searches affect another's ads. Under that model, repeated checks by activists could turn an accusation of bias into part of the signal that produces it.

Enforcement economics: the fine is small relative to the company

The recurring complaint in the thread is that the penalty is not a deterrent. The daily fine of R$500,000, about $99,000, and the threatened doubling to about $198,000, were repeatedly described as negligible for a company of Meta's size.

The daily fine is so meaninglessly small ($99,000 per day) that there is very little incentive to do anything.

Another commenter put the same point more bluntly:

This will only get fixed when people get jailed by ignoring court orders, fines for big tech do nothing.

A third framed it as a rational calculation:

There's no downside really to ignoring the courts order.

Brazilian electoral process context from a local observer

A commenter who described living in Brazil outlined how the TSE process normally works: shortly before campaigns begin, the electoral court meets with representatives of the social platforms operating in Brazil to discuss the rules of the game and the algorithms; by the time campaigns are underway, rules are set and agreements have been signed; monitoring continues throughout the campaign. In that account, the pre-campaign rules plus observed results lead to a kind of fine-tuning. The same commenter offered a skeptical reading of the timing of the latest demands, saying the last-minute pressure may be intended to extract concessions or disclose algorithmic information to the Brazilian state rather than purely to correct results.

The sovereignty argument

A separate line of argument in the thread treats the dispute as a question of jurisdiction rather than of specific GIFs. One commenter argued that when Brazilian law is applied to these companies, they appeal to the United States president and the issue becomes diplomatic, and concluded that countries need domestic platforms that abide by local law. Another wrote that American billionaires are deciding politics in the United States, South America and Europe, and predicted that the resulting resentment will produce conflicts or regulation.

Others in the thread pointed to X as the precursor case, citing reporting that hard-right content is promoted and protected from content-violation enforcement on that platform, and describing it as the first experiment in what electioneering looks like when large platforms are not constrained.

Counterpoints from the thread about the courts and about Meta's record

Not all skepticism in the discussion aimed at the petitioners. Some commenters questioned the neutrality of Brazilian courts, noting that the Supreme Federal Court has been criticized for censorship and citing a New York Times finding of 223 account-blocking orders since 2020. Another commenter countered the favoritism narrative by pointing to the fact that Meta and other large platforms had previously banned the sitting president of the United States.

A Brazilian commenter disputed the premise of the story directly, saying that the Workers' Party is a target of hatred on social media and that its communication is inefficient, and describing the situation as the outcome of a party that has been in power for 20 years. Another commenter gave anecdotal color, saying the Brazilians they know in the United States are voting for Bolsonaro.

A different objection noted the familiar reference point of the Google hearing in which a questioner asked why a search for "idiot" returned a political figure's image.

The protest: "Meta, Hands Off Our Election"

Besides the legal action, Sleeping Giants Brasil and CTRL+Z planned a demonstration outside Meta's headquarters in São Paulo at 5 p.m. Monday. The event is called "Meta, Hands Off Our Election," and the organizations say it will be peaceful. Participants plan to hand a notice to the company's director in Brazil and want action on the suspicions of unequal treatment of the candidacies. The mobilization puts pressure on Meta in court and in public debate at the same time.

What is established and what is not

Established by the reporting: a petition was filed with the TSE by Sleeping Giants Brasil and CTRL+Z; it names Justice Marco Aurélio Bellizze as the case's overseer; it cites specific GIF, hashtag and Meta AI outputs; the court has already ordered equal treatment and a proof of compliance; the full court upheld that order; a new 24-hour deadline was set; and a daily fine of R$500,000, with a threat to raise it to R$1 million, is on the table. Meta has denied deliberate favoritism and attributed earlier problems to technical failures.

Not established: that Meta intentionally favored Flávio Bolsonaro. The organizations filing the petition explicitly pose the ranking question to the court rather than claiming to have proven bias, and the Meta AI number coincidence is presented by the source as insufficient on its own to conclude electoral preference. The Hacker News discussion adds a competing technical explanation — engagement-driven ranking and personalized results — that would produce similar outputs without any political intent in the system.