{"version":"1.9","identifier":"8d59fad9-13c0-4f6d-8365-63733b84c552","language":"en-US","title":"Brazil fines Meta R$7.5m over runoff orders it only partly obeyed, but finds no proof it tried to tip the vote","layout":{"columns":7,"width":1024,"margin":32,"gutter":20},"componentLayouts":{"text":{"columnStart":0,"columnSpan":7,"maximumContentWidth":720,"margin":{"top":12,"bottom":12}},"photo":{"columnStart":0,"columnSpan":7,"margin":{"top":20,"bottom":20}}},"componentTextStyles":{"default":{"fontName":"HelveticaNeue","fontSize":18,"lineHeight":29,"textColor":"#171717"},"headline":{"fontName":"HelveticaNeue-Bold","fontSize":36,"lineHeight":42,"conditional":[{"conditions":{"maxViewportWidth":480},"fontSize":28,"lineHeight":34}]},"intro":{"fontName":"HelveticaNeue","fontSize":20,"lineHeight":28,"textColor":"#555555"},"caption":{"fontName":"HelveticaNeue","fontSize":13,"lineHeight":19,"textColor":"#666666"},"byline":{"fontName":"HelveticaNeue","fontSize":13,"lineHeight":20,"textColor":"#666666"},"body":{"fontName":"HelveticaNeue","fontSize":18,"lineHeight":29,"paragraphSpacingAfter":14,"linkStyle":{"textColor":"#00704a"}}},"metadata":{"canonicalURL":"https://567politics.com/news/brazil-tse-fines-meta-7-5-million-runoff-orders-bellizze-meta-ai-november-2026","datePublished":"2026-11-11T12:03:35.508967Z","dateModified":"2026-11-11T12:06:41.128391Z","authors":["567 Politics staff"],"excerpt":"Brazil's electoral court says Meta missed deadlines and gave vague answers. Meta AI refused 98.1% of Lula image prompts vs 24.7% for Flávio before a fix. An abuse probe goes on.","thumbnailURL":"https://pub-fac99da959e344fe99d810d309cef7df.r2.dev/photos/article/8d59fad9-13c0-4f6d-8365-63733b84c552/1c3af7c2d101ba5caeb6464cf82a28a9.jpg","keywords":["br"],"generatorName":"Frog News","generatorVersion":"1.0"},"components":[{"role":"title","text":"Brazil fines Meta R$7.5m over runoff orders it only partly obeyed, but finds no proof it tried to tip the vote","textStyle":"headline","layout":"text"},{"role":"intro","text":"Brazil's electoral court says Meta missed deadlines and gave vague answers. Meta AI refused 98.1% of Lula image prompts vs 24.7% for Flávio before a fix. An abuse probe goes on.","textStyle":"intro","layout":"text"},{"role":"byline","text":"By 567 Politics staff · Published November 11, 2026","textStyle":"byline","layout":"text"},{"role":"photo","URL":"https://pub-fac99da959e344fe99d810d309cef7df.r2.dev/photos/article/8d59fad9-13c0-4f6d-8365-63733b84c552/1c3af7c2d101ba5caeb6464cf82a28a9.jpg","caption":"The Superior Electoral Court (TSE) headquarters in Brasília, designed by Oscar Niemeyer. Photo: Luiz Roberto/TSE, released under the Public Domain Mark. · Luiz Roberto/TSE via Wikimedia Commons (Public Domain Mark)","accessibilityCaption":"The Superior Electoral Court (TSE) headquarters in Brasília, designed by Oscar Niemeyer. Photo: Luiz Roberto/TSE, released under the Public Domain Mark.","layout":"photo"},{"role":"caption","text":"The Superior Electoral Court (TSE) headquarters in Brasília, designed by Oscar Niemeyer. Photo: Luiz Roberto/TSE, released under the Public Domain Mark. · Luiz Roberto/TSE via Wikimedia Commons (Public Domain Mark)","textStyle":"caption","layout":"text"},{"role":"body","format":"markdown","text":"Brazil's Superior Electoral Court (TSE) fined Meta R$7.5 million on Tuesday, November 10. The vote was unanimous. The court said Meta failed to fully obey orders it issued during the presidential runoff to guarantee equal treatment of [Luiz Inácio Lula da Silva](https://567politics.com/politicians/luiz-inacio-lula-da-silva) and [Flávio Bolsonaro](https://567politics.com/politicians/flavio-bolsonaro) on Facebook, Instagram and WhatsApp. It did not find that Meta tried to tip the election. Beyond technical failures the company itself admitted, the record shows no \"deliberate action to unbalance the election,\" wrote the rapporteur, Justice Marco Aurélio Bellizze. Such a finding would need \"robust proof\" in a fuller proceeding, according to [his written vote, published by Migalhas](https://arq.migalhas.com.br/arquivos/2026/11/af1a0a53_tse-rp-0602746-69-voto-ministro-bellizze.pdf).\n\nThe fine punishes conduct toward the court, not conduct toward voters. Bellizze listed three reasons for it: Meta replaced the court's deadlines with its own, delivered information late and in pieces, and answered several orders with generic declarations, aggregated data or documents in English. All six other justices followed him without reservations, [Folha de S.Paulo reported](https://www1.folha.uol.com.br/poder/2026/11/relator-no-tse-vota-para-condenar-meta-a-multa-de-r-75-mi-por-tratamento-desigual-entre-lula-e-flavio-nas-eleicoes.shtml). Meta said in a statement that it \"complies with Brazilian electoral law and applies the same rules to all candidates and parties\" and will keep defending itself \"against false allegations\" in pending cases.\n\nThe ruling is also the most detailed public account so far of what went wrong on Meta's platforms in the weeks before the [October 25 runoff](https://567politics.com/news/brazil-runoff-lula-wins-fourth-term-polls-extension-october-2026), which Lula won. One figure stands out. Meta's own technical statement, quoted in the vote, covers image requests naming a candidate and his ballot number. Before an October 14 fix, its Meta AI tool refused 98.1% of such requests about Lula, against 24.7% for Flávio Bolsonaro. A separate investigation into whether the platforms' conduct amounted to election abuse is still open.\n\n## What the court decided, and what it did not\n\nThe case is a \"representação\" under article 96 of Brazil's Elections Law (Law 9.504/1997), filed by Lula's coalition, Brasil Pronto pra Mais, against Facebook Serviços Online do Brasil Ltda., Meta's Brazilian unit. Flávio Bolsonaro joined as an interested third party. The vote describes the procedure as summary. The defendant has 48 hours to answer and the decision is due within 24 hours. There is \"no room for expert examination, complex technical evidence or broad evidentiary discovery.\"\n\nThat limit shapes everything else in the ruling. On October 15 Bellizze granted a partial injunction with 17 orders, labeled 'a' to 'q'. Each had its own deadline, backed by a daily fine of R$500,000. The full court upheld it unanimously on October 16. On October 18 Bellizze gave Meta 24 hours to prove full compliance with 11 pending items, or the daily fine would double to R$1 million. He also required all compliance documents in Portuguese, with an explanation.\n\nOn Tuesday the court:\n\n* confirmed the injunction, which expired with the runoff on October 25;\n* found partial noncompliance and fixed the fine at R$7.5 million;\n* rejected Flávio Bolsonaro's motion for clarification;\n* refused the coalition's requests to raise the daily fine to R$10 million, to find Meta in contempt (\"ato atentatório à dignidade da justiça\") and to send Meta's documents to the Federal Police for technical analysis;\n* declined to consider a filing by the civil-society groups Sleeping Giants Brasil and Ctrl+Z, which are not parties.\n\nThe vote is explicit that the fine is \"coercive.\" It is \"not related to any intent to favor a given candidacy.\" Bellizze rejected Meta's argument that the case became moot when the election ended, because the subject is compliance with court orders. On contempt, he wrote that the record shows late and incomplete compliance. It does not allow a finding of \"deliberate resistance\" to the court.\n\n### How R$7.5 million was calculated\n\nThe amount is the daily fine applied to nine days. The vote counts three days at R$500,000, from October 17 to 19, and six days at R$1 million, from October 20 to the runoff on October 25. Bellizze wrote that the figure reflects Meta's conduct \"as a whole,\" not a sum of orders met or missed, and is \"not disproportionate to the economic capacity\" of the company.\n\n**Daily fine the TSE counted against Meta, October 2026**\n\nMillions of reais per day; total R$7.5 million. The R$1 million rate applied after Meta missed the 24-hour deadline set on October 18.","textStyle":"body","layout":"text"},{"role":"htmltable","html":"<table><tr><th>Day</th><th>Daily fine</th></tr><tr><td>Oct 17</td><td>R$0.5m</td></tr><tr><td>Oct 18</td><td>R$0.5m</td></tr><tr><td>Oct 19</td><td>R$0.5m</td></tr><tr><td>Oct 20</td><td>R$1m</td></tr><tr><td>Oct 21</td><td>R$1m</td></tr><tr><td>Oct 22</td><td>R$1m</td></tr><tr><td>Oct 23</td><td>R$1m</td></tr><tr><td>Oct 24</td><td>R$1m</td></tr><tr><td>Oct 25</td><td>R$1m</td></tr></table>","layout":"text"},{"role":"body","format":"markdown","text":"Source: [Rapporteur's vote, Rp 0602746-69.2026.6.00.0000 (Migalhas)](https://arq.migalhas.com.br/arquivos/2026/11/af1a0a53_tse-rp-0602746-69-voto-ministro-bellizze.pdf)\n\n## Order by order: what Meta met and what it missed\n\nThe vote goes through each order. It is more granular than the headline fine suggests, and it credits Meta on several points.\n\n**Orders the court considered met.** Meta confirmed that no \"non-recommendation\" measure applied to accounts other than the candidates' official profiles (order 'b'). Brazilian electoral rules bar platforms from algorithmically recommending candidates' own profiles. Joint \"collab\" posts posed a technical problem. Meta said it could not stop recommending such a post from the official profile while still recommending it from a co-author's profile. It kept the whole post out of recommendations. Bellizze accepted that, because it applied identically to both campaigns. Meta also reversed a September 23 restriction on Lula's campaign ad account, which its own report said lasted about an hour (order 'k'). And it showed that every ad bought on September 23 and October 9 and 10 was eventually delivered (order 'l'). The coalition objected that it could not create new ads during the October 9–10 failure. The court said that is different from paid delivery that never happened, and the coalition named no pending delivery.\n\n**Orders met late.** Meta's allowlist limiting election ads to candidates, parties, federations and coalitions took effect only at midnight on October 20. That day it paused 7,068 ads (order 'd'). Meta also reviewed 255 Instagram profiles that the coalition called an AI-run network. Its report, quoted in the ruling, said 27 accounts were disabled, five faced identity checks and one had been deactivated by its user. Another 188 of the accounts examined showed no violations (order 'g'). Information about third-party accounts arrived on October 22 (order 'm').\n\n**Orders met only in part.** Meta said it was not reducing distribution of either campaign's content (order 'a'). The court found that assurance rested on \"generic declarations of equality.\" The Electoral Prosecutor's Office (PGE) had flagged a translation tool that, in a controlled test, flipped the meaning of comments opposing Flávio Bolsonaro. It rendered \"Flávio futuro presidiário\" (\"Flávio, future prisoner\") as \"Flávio future president.\" Similar phrases without his name were translated correctly. Meta's report on the 7,068 suspended ads (order 'e') was also incomplete. About 1,530 records, a little over a fifth, lacked the ad library ID, run dates, reach, spending, buyer and country. Meta did set up human review of automated restrictions on campaign ad accounts (order 'c'). But it reported only one such case to the court and never said whether there were others.\n\n**Orders not met.** The court said Meta never delivered the 48-hour report, required under order 'q', describing the measures it had actually taken. The consolidated report it filed after the runoff, on October 30, labeled every order as complied with but, the vote says, added no new evidence. Delivery series and internal tickets for the October 7 and October 9–10 incidents (part of order 'n') were never filed; Meta told the court it was \"working tirelessly\" to provide them.\n\nA late filing did not change the result. On November 6 Meta submitted a sealed technical note on Instagram distribution from August 16 to October 25. Bellizze quoted the note's own caveat. Its figures \"reflect both what the audience chose to watch and what the platform delivered, and do not evaluate how ranking treated each account.\" So, he concluded, it could not show that identical criteria were applied.\n\n## The Meta AI numbers\n\nThe most concrete evidence of unequal treatment came from Meta itself. The coalition said its tests on October 13 showed Meta AI refusing election-themed images of Lula. The same tool produced campaign-style material favorable to his opponent, it said. Later the coalition told the court of a new episode it learned of on October 19. Inside WhatsApp, the tool answered neutral prompts about \"president\" or \"elections\" with posters reading \"Vote 22,\" Flávio Bolsonaro's ballot number. It refused the same request with Lula's number, 13.\n\nMeta's technical statement for order 'o' acknowledged that, until additional measures on October 14, the tool did not recognize Lula's opponent by his first name. Requests naming that candidate and his ballot number were refused 24.7% of the time; equivalent requests about Lula were refused 98.1% of the time.\n\n**Meta AI image requests refused before Meta's October 14 fix**\n\nRequests naming each candidate with his ballot number, before Meta's October 14 measures. Meta's own sealed statement, as quoted in the ruling; its statistics cover October 13–15, 2026 only.","textStyle":"body","layout":"text"},{"role":"htmltable","html":"<table><tr><th>Candidate named</th><th>Refused</th></tr><tr><td>Lula (PT, 13)</td><td>98.1%</td></tr><tr><td>Flávio Bolsonaro (PL, 22)</td><td>24.7%</td></tr></table>","layout":"text"},{"role":"body","format":"markdown","text":"Source: [Rapporteur's vote, Rp 0602746-69.2026.6.00.0000 (Migalhas)](https://arq.migalhas.com.br/arquivos/2026/11/af1a0a53_tse-rp-0602746-69-voto-ministro-bellizze.pdf)\n\nBellizze drew a narrow conclusion. The data \"indicate different treatment between the candidacies\" before October 14, and the fix \"reduced the asymmetry.\" But Meta did not prove the tool applied identical criteria for the rest of the campaign. The PGE noted that the \"Vote 22\" episode came after Meta had told the court the flaw was fixed. The vote does not say why the tool behaved this way. Meta's filter documents, including the versions in force on October 13 and 14, remain sealed as trade secrets, and the court never received the stress-test results Meta said it had run.\n\nThose numbers do not show that Meta AI changed votes, and the ruling does not claim they did. No evidence in the public record measures how many voters used the tool, what they saw or whether it affected their choices.\n\n## Synthetic content and paid ads\n\nTwo other strands concerned rules specific to Brazilian elections. TSE Resolution 23.610/2019 requires labels on AI-generated or \"synthetic\" political content. The court's order 'f' also required Meta to block new synthetic content covered by the rule from 72 hours before to 24 hours after the runoff. Meta reviewed 1,204 posts the coalition had flagged. Of those, 270 were already labeled, 720 were made unavailable and 214 were classified by Meta as not electoral. The court accepted that review. The coalition said Meta's classification of the 214 could not be checked.\n\n**What Meta did with 1,204 posts flagged for missing AI labels**\n\nMeta's own classification, accepted by the court under order 'h'.","textStyle":"body","layout":"text"},{"role":"htmltable","html":"<table><tr><th>Outcome</th><th>Posts</th></tr><tr><td>Made unavailable</td><td>720</td></tr><tr><td>Already labeled</td><td>270</td></tr><tr><td>Classed as not electoral</td><td>214</td></tr></table>","layout":"text"},{"role":"body","format":"markdown","text":"Source: [Rapporteur's vote, Rp 0602746-69.2026.6.00.0000 (Migalhas)](https://arq.migalhas.com.br/arquivos/2026/11/af1a0a53_tse-rp-0602746-69-voto-ministro-bellizze.pdf)\n\nMeta proved it had configured the blackout window for new synthetic content from October 22 to 26. It did not prove it had switched off paid boosting during the legal blackout, or fixed its labeling mechanism, both required by October 21.\n\nThe paid-ad allowlist matters for a simple reason. Meta was close to the only seller of online political ads in 2026, since Google stopped selling them in 2024, [Folha reported in October](https://www1.folha.uol.com.br/poder/2026/10/em-mensagens-internas-meta-admite-que-contas-de-lula-foram-prejudicadas-por-restricoes-engenheiros-falam-em-falhas-tecnicas.shtml). TSE rules bar algorithmic recommendation of candidates' accounts, which may appear in search only when a user types the exact name. So paid boosting is the only way for candidates to widen their reach, Folha noted. Flávio Bolsonaro's lawyers complained to the court about the same ad pipeline on October 16, from the other direction. They said his campaign suffered \"on a much larger scale\" from paid negative ads. Some were boosted by Lula's official page, they said, and others by ghost pages, individuals and organizations with no right to buy election ads. They called Meta \"negligent,\" [g1 reported](https://g1.globo.com/politica/noticia/2026/10/17/depois-de-lula-flavio-bolsonaro-diz-ser-prejudicado-pela-meta-e-pede-para-tse-proibir-impulsionamento-de-posts-negativos.ghtml). The court admitted him to the case. It did not take up his requests for new orders, which it said belonged in a separate action.\n\n**The TSE's case against Meta, from first failures to the fine**\n\n* **2026-09-23**: Lula campaign ad account restricted for about an hour by an automated sanction\n* **2026-10-07**: Global view-count outage; Lula video shows more likes than views\n* **2026-10-09**: Lula's official Instagram cannot boost feed posts for nearly a day\n* **2026-10-13**: Coalition says its tests show Meta AI refusing Lula images\n* **2026-10-14**: Meta adds filters for candidates' first names and ballot numbers\n* **2026-10-15**: Bellizze issues 17 orders, R$500,000 daily fine\n* **2026-10-16**: Full court upholds the orders unanimously\n* **2026-10-18**: Case-management order; fine rises to R$1 million a day after 24 hours\n* **2026-10-20**: Meta's allowlist takes effect; 7,068 ads paused\n* **2026-10-22**: Corregedor opens abuse probe (AIJE) and orders Meta and X to explain\n* **2026-10-25**: Runoff; injunction expires\n* **2026-10-30**: Meta files consolidated report calling every order met\n* **2026-11-06**: Meta files sealed Instagram distribution note\n* **2026-11-10**: TSE fines Meta R$7.5 million, unanimously\n\nSource: [Rapporteur's vote (Migalhas)](https://arq.migalhas.com.br/arquivos/2026/11/af1a0a53_tse-rp-0602746-69-voto-ministro-bellizze.pdf), [TSE](https://www.tse.jus.br/comunicacao/noticias/2026/Outubro/corregedor-do-tse-da-24-horas-para-meta-e-x-explicarem-possivel-tratamento-desigual-de-candidaturas-nas-redes)\n\n## The competing cases\n\n**Lula's coalition** argues that the ruling confirms a problem bigger than a fine. Its lawyer, Angelo Ferraro, told the court \"Meta's black box needs to be opened\" and \"victory at the polls does not absolve Meta,\" [CNN Brasil reported](https://www.cnnbrasil.com.br/politica/tse-meta-julgamento-eleicoes/). He said that \"in this election Meta stopped being a platform and became a political agent,\" and that Brazil's president \"will not be chosen by Silicon Valley or Washington,\" according to [g1](https://g1.globo.com/politica/noticia/2026/11/10/tse-meta-descumprir-decisao-sobre-eleicoes.ghtml). His central argument, reported by [Estadão](https://www.estadao.com.br/politica/tse-multa-meta-descumprir-decisoes-de-isonomia-entre-lula-e-flavio/), is that a penalty should not require proof of intent: \"For Meta, all of this was an error. But what it doesn't explain is that the errors were only to one side. And an error, in continuity, is not just an error, it is a pattern.\" The campaign moved to seek a broader investigation, [Folha reported](https://www1.folha.uol.com.br/poder/2026/10/campanha-de-lula-deve-pedir-investigacao-contra-meta-apos-mensagens-internas-da-empresa.shtml), after Folha published internal Meta messages on [October 16](https://www1.folha.uol.com.br/poder/2026/10/em-mensagens-internas-meta-admite-que-contas-de-lula-foram-prejudicadas-por-restricoes-engenheiros-falam-em-falhas-tecnicas.shtml). In them, employees acknowledged that Lula's Instagram account was \"limited\" compared with Flávio Bolsonaro's and that Flávio Bolsonaro's profile had been wrongly recommended in Instagram search.\n\n**Meta** argues that it complied and did not take sides. Its lawyer, Patrícia Helena Marta Martins, told the court that claims Meta favored any candidate are \"false and unfounded.\" She said all orders were met and the failures were brief technical problems that had been explained, [CNN Brasil reported](https://www.cnnbrasil.com.br/politica/tse-meta-julgamento-eleicoes/). In an October filing quoted by [CNN Brasil](https://www.cnnbrasil.com.br/eleicoes/meta-reconhece-erro-em-anuncios-de-lula-mas-nega-vies-no-algoritmo/), Meta said its systems \"do not operate according to ideological criteria.\" It cited its record in \"more than 200 elections since 2016.\" Meta told the TSE that the October 7 outage was global, [Estadão reported](https://www.estadao.com.br/politica/eleicoes/meta-nega-favorecimento-a-campanha-de-flavio-em-resposta-ao-tse/). It said the Lula ad-account suspension came from fraud-detection measures it had tightened under an agreement with Brazil's Attorney General's Office (AGU). In a November 6 filing in a separate abuse probe, Meta said Lula's Instagram accounts reached about 1.5 times as many people as Flávio Bolsonaro's. That figure covers organic distribution from August 16 to October 25, [O Tempo reported](https://www.otempo.com.br/politica/judiciario/2026/11/6/meta-usa-dados-sigilosos-para-negar-ao-tse-favorecimento-a-flavio-e-tenta-barrar-pericia).\n\n**Flávio Bolsonaro's campaign** argues that equality means the same rules, not the same results, and that the record ignores harms to his side. His filings are summarized in the vote. He said the orders should apply to his candidacy too, and that data on his official profile were requested without his participation. He said he was harmed by negative ads boosted by Lula's campaign and by buyers with no legal right to buy them. He also argued that the Sleeping Giants filing relied on new facts that would need expert evidence. The court rejected his clarification motion, saying it sought to reopen what had been decided.\n\n**The court** placed itself between these positions. Bellizze wrote that platforms hold \"informational power\" exercised through \"opaque algorithmic systems.\" He called for \"effective mechanisms of algorithmic governance, auditing and oversight.\" He also wrote that the case involved \"no undue intervention by the State in the private sphere.\" Vice-president André Mendonça said during the session that he had perceived \"a certain resistance on the part of the company\" to complying, [g1 reported](https://g1.globo.com/politica/noticia/2026/11/10/tse-meta-descumprir-decisao-sobre-eleicoes.ghtml).\n\n**Critics of court-led internet rules**, a strand found largely on the right, argue that the remedy should come from Congress, not from the electoral courts. Francieli Campos, a specialist in electoral and digital law, made that case in a December 2025 interview with [Gazeta do Povo](https://www.gazetadopovo.com.br/vida-e-cidadania/algoritmos-vao-ser-o-juiz-das-redes-no-lugar-do-stf-em-2026-alerta-jurista/). She said Brazil's election rules are \"among the most restrictive in the world\" and that the electoral courts \"also legislate\" through resolutions. She warned that the Supreme Court's 2025 reinterpretation of platform liability risks \"automated prior censorship.\" That interview predates this case and does not address the Meta ruling. The sources reviewed for this article contained no public reaction to Tuesday's decision from Flávio Bolsonaro or PL leaders.\n\n**Civil-society groups** that backed the complaint want legislation. On November 10, a coalition including Ctrl+Z, Sleeping Giants Brasil, Coding Rights and the National Forum for the Democratization of Communication (FNDC) published a letter to Lula. It calls for platform rules built around \"transparency, due process and oversight,\" [FNDC said](https://fndc.org.br/censura-privada-carta-a-lula-cobra-responsabilidade-das-big-techs/). The proposals include notice and appeal channels for moderation decisions and stronger powers for the National Data Protection Authority (ANPD). The groups said they would propose a provisional measure. They also urged passage of a digital-markets competition bill, PL 4.675/2025, and the AI bill, PL 2.338/2023. A Big Tech competition bill is already on [Lula's short year-end agenda](https://567politics.com/news/brazil-lula-year-end-agenda-bets-desenrola-big-tech-november-2026).\n\n## Weighing the arguments\n\nEach side's strongest claim has a gap that the public record does not close.\n\nThe coalition's \"pattern\" argument asks the court to infer bias from the direction of several errors. Direction matters, and Meta's own Meta AI figures do show one-sided treatment for a period. But a run of errors that hurt one side is consistent with more than one explanation, including bias, differences in how each campaign used the platforms, or flaws that hit some accounts harder. Treating the sequence alone as proof of intent would be a hasty generalization. Bellizze's answer was that intent cannot be tested in a summary procedure without the technical evidence Meta controls.\n\nMeta's defenses have their own logical limits. A record across \"more than 200 elections\" does not show what happened in this one; it is an appeal to past conduct, not evidence about October 2026. Its total-reach comparison answers a different question from the one the coalition asked. As O Tempo noted, a larger total audience can come from posting more often even if each post was distributed more narrowly. Meta's own note told the court its aggregate figures \"do not evaluate how ranking treated each account.\"\n\nThe court's position also rests on a disputed premise: that judges can and should demand algorithmic transparency during an election. Critics such as Campos argue that the electoral courts already write rules Congress has not passed. Supporters of the ruling reply that the court enforced rules already on the books, including article 9-D of Resolution 23.610, which imposes a \"duty of care\" on platforms. That is a value disagreement about institutional roles more than a factual one.\n\n## The abuse probe and the documents still under seal\n\nThe fine does not end the legal fight. The TSE will keep examining whether Meta favored any candidacy, with no date for a final judgment, Agência Brasil [reported](https://agenciabrasil.ebc.com.br/justica/noticia/2026-11/tse-multa-meta-em-r-75-milhoes-por-descumprir-decisao-nas-eleicoes). The main open venue for that question is an electoral-abuse investigation, AIJE 0603276-73.2026.6.00.0000. Lula's coalition filed it against Flávio Bolsonaro and his running mate, Alfredo Gaspar de Mendonça Neto, alleging abuse of economic power and misuse of the media. On October 22 the corregedor-geral, Justice Ricardo Villas Bôas Cueva, gave Meta and X 24 hours to explain how they handled political content, AI tools and election ads, [the TSE said](https://www.tse.jus.br/comunicacao/noticias/2026/Outubro/corregedor-do-tse-da-24-horas-para-meta-e-x-explicarem-possivel-tratamento-desigual-de-candidaturas-nas-redes). Cueva stressed that opening the investigation \"does not mean recognition of the accusations.\" Meta is not a defendant in that action.\n\nThe stakes there are different. Brazil's Ineligibility Law allows a court that finds abuse in an AIJE to annul a candidate's registration or diploma and bar the candidate for eight years, as [O Tempo noted](https://www.otempo.com.br/politica/judiciario/2026/11/6/meta-usa-dados-sigilosos-para-negar-ao-tse-favorecimento-a-flavio-e-tenta-barrar-pericia). Flávio Bolsonaro lost the runoff and has [since left Brazil](https://567politics.com/news/brazil-flavio-bolsonaro-leaves-after-runoff-october-2026). In that case the coalition wants an expert audit of Meta's recommendation systems and testimony from employees. Meta opposes both, saying an audit would expose trade secrets without changing the conclusion its records support.\n\nNone of the reports reviewed for this article said whether Meta would appeal Tuesday's fine. Under the ruling, Lula's coalition gets access to Meta's compliance documents except the Meta AI filters and synthetic-content detection parameters, which stay sealed as trade secrets. Flávio Bolsonaro does not get access to those either.\n\nThe ruling comes amid other moves on online speech in Brasília. This week the Supreme Court closed the long-running [fake news inquiry](https://567politics.com/news/brazil-stf-fachin-archives-fake-news-inquiry-4781-moraes-november-2026). Civil-society groups say Lula acknowledged after his victory that platform rules are urgent. Any such law would need votes from the next Congress, where the [Liberal Party (PL)](https://567politics.com/parties/brazil/pl) will hold the largest bench from February, not just from Lula's [Workers' Party (PT)](https://567politics.com/parties/brazil/pt) and its allies.\n\n## What is known, and what is not\n\nKnown, from the court record: Meta missed deadlines on several orders and filed some material in English or in aggregate form. By its own figures, its AI tool refused Lula-related image requests far more often than Flávio-related ones until its October 14 fix. The court found these failures serious enough to fine the company. It found no evidence, in this procedure, that Meta acted deliberately to favor a candidate.\n\nNot known: whether Meta's ranking systems treated the two campaigns differently overall, and why the Meta AI asymmetry arose. Also unknown is how many voters were exposed to any of these failures, and whether any of it affected the [2026 presidential result](https://567politics.com/elections/brazil/brazil-president-2026). The technical documents that might answer those questions are sealed, and the abuse investigation has not ruled.\n\nNo representative poll on Brazilians' views of Meta's conduct in the election, or of platform regulation after the runoff, was found for this article. Statements by campaigns, the company and advocacy groups should not be read as a measure of public opinion.\n\n## Sources and method\n\nThis article is based on the 47-page written vote of Justice Bellizze in Representação 0602746-69.2026.6.00.0000, published by Migalhas. It also draws on reporting by g1, CNN Brasil, Folha de S.Paulo, Estadão, O Globo, Valor Econômico, Agência Brasil, Gazeta do Povo and O Tempo. Other sources are a TSE press release and the FNDC's publication of the civil-society letter. Figures about Meta's internal data come from Meta's own filings as quoted by the court; 567politics has not seen the sealed documents. Brazilian campaign-finance disclosures were not relevant to the court's findings and were not used. Portuguese quotations were translated by 567politics.","textStyle":"body","layout":"text"}]}