Meta AI, Lula images and Brazil's election: what's reported, and what Meta and Brazil's electoral court say

The dispute over Meta AI and Lula images reached Brazil's top electoral court in the final weeks of the 2026 presidential race. President Luiz Inácio Lula da Silva's campaign says Meta's AI assistant refused prompts for pro-Lula campaign images while producing them for his opponent, Senator Flávio Bolsonaro. On October 15, 2026, the court ordered Meta to treat both candidates equally. Meta denies favoring anyone. After a first round on October 4, Lula won the runoff on Sunday, October 25, 2026, with 51.8% of the vote and a lead of nearly 4.4 million votes, according to the electoral court's full count as reported by the AP. As of October 26, no ruling on Meta's compliance or on the court's investigation had been reported.

This story is based on releases from Brazil's Superior Electoral Court (TSE), reporting by Agência Brasil, ConJur, Poder360 and Gazeta do Povo, and reporting by the AP, Reuters and Bloomberg, checked October 26, 2026. Portuguese quotes are 9to5AI's translations. Allegations by the campaigns and prosecutors haven't been ruled on. For how the US handles AI in political ads, see our guide to AI political ad labels.

The short version

• The claim: Lula's campaign told the TSE that Meta AI blocked image prompts about Lula but allowed similar ones about Flávio Bolsonaro, and that Meta limited the reach of Lula's posts.
• October 15: Justice Marco Aurélio Bellizze ordered equal treatment, including identical criteria in Meta AI, under a R$500,000 daily fine. He found the tool had failed but said there was no evidence the failure was deliberate.
• Meta's answer: technical problems, not bias. "Any suggestion that Meta favors one side is false," it said, as Reuters reported.
• October 21: electoral prosecutors told the TSE that Meta hadn't fully complied. Meta denied it.
• October 22: the TSE's inspector-general gave Meta and X 24 hours to explain how their AI and feeds treat political content.
• October 25: Lula won the runoff. No ruling on the fine or the investigation had been reported as of October 26.

What did Lula's campaign say Meta AI did?

The AP reported on October 15 that the campaign's complaint said users couldn't generate campaign images of Lula through the Meta AI feature in WhatsApp, but could generate images promoting Flávio Bolsonaro. ConJur reported that the campaign's evidence showed the image tool blocking pro-Lula prompts while allowing similar images for his rival. Bloomberg, in an October 24 story by Augusta Saraiva relayed by Yahoo Finance, reported that the campaign accuses Meta of "repeatedly and systematically" failing to deliver Lula's content.

The complaint goes beyond AI. The campaign says some of Lula's Instagram posts appeared without a share button or with reduced reach. On October 16, Reuters, citing Folha de S.Paulo, reported internal Meta messages in which employees acknowledged Lula's Instagram accounts had been negatively affected by restrictions, and that the platform had recommended Flávio Bolsonaro's profile and WhatsApp channel. 9to5AI hasn't seen those messages.

What did Brazil's electoral court order?

Last verified: October 26, 2026

Bellizze's first order, as reported by Agência Brasil, found a failure in the tool that controls prompts. But he wrote that there were no elements "that allow one to say that this failure resulted from a deliberate criterion." The order doesn't cover opinions posted by users.

The October 22 investigation was filed by Lula's coalition, "Brasil Pronto pra Mais," against the Flávio Bolsonaro ticket. It alleges abuse of economic power and misuse of the media. The TSE said it also asked X to explain how its Grok assistant works, and ordered a human review of Meta's removal of a video it had taken down citing AI use. The Rio Times reported that the judge said opening the case doesn't mean the allegations are true.

What does Meta say?

Meta's position across filings and statements is that the problems were technical:

• To Reuters, October 16: the issues were already public and internal conversations concerned their severity, not political interference. "Any suggestion that Meta favors one side is false."
• To the TSE, by Agência Brasil's account: Meta said it had "temporarily" stopped Meta AI from recommending the candidates or generating images of them, and blocked prompts that used only a candidate's first name and ballot number.
• At the October 20 hearing: Meta "denied any interference in the electoral process or political favoritism" and attributed the problems to "technical issues," the TSE said.
• On compliance: Meta told the court it is complying in full while reserving the right to seek revisions, Gazeta do Povo reported.

Did Meta ignore the court's order?

That's contested. On October 21, Brazil's electoral prosecutors (the Ministério Público Eleitoral) told the TSE that Meta hadn't fully complied and asked for the fine to be applied, Gazeta do Povo reported. They said Meta's AI still gave different answers depending on the candidate. Lula's campaign added that generic prompts about the presidency produced images showing the number 22, Flávio Bolsonaro's ballot number. Meta denied any breach. As of October 26, 2026, the day after the runoff, no report showed Bellizze had ruled on the fine. Claims online that Meta "ignored" the order go further than the record supports.

What happened in the runoff?

Lula won a fourth term on October 25, 2026, with 51.8% of the vote, the AP reported, after trailing Flávio Bolsonaro 47% to 45% in the first round. Flávio Bolsonaro refused to concede and said he would ask the TSE to annul the result. His stated grounds, as InfoMoney reported on October 25, were a court decision extending voting hours and his party's poll watchers being removed from the counting room; he didn't cite Meta or AI. The Meta order applied until the end of the runoff, but the fine request and the inspector-general's investigation are separate matters that the vote doesn't settle.

Why does it matter?

The case puts a court order on how a generative AI tool must treat candidates, not just on posts or ads. Brazil's rules also bar new synthetic content and paid promotion of campaign material from October 22, 2026, Agência Brasil noted. In the US, where the midterms are on November 3, 2026, there's no federal rule of that kind, as our guide to AI political ad labels explains. For what Meta AI is built on, see our Muse Spark explainer; for another fight over Meta's terms, see our story on its October 30 commercial terms.

What's confirmed, reported and uncertain

• Confirmed (TSE releases): the October 20 hearing and Meta's stated position there; the October 22 investigation, its case number, parties and R$500,000 fine; the first-round and runoff dates of October 4 and October 25, 2026 (TSE calendar, via O Liberal).
• Reported: the October 15 order's terms and Bellizze's finding (Agência Brasil, AP, Reuters); the October 16 majority and October 18 deadline (ConJur, Poder360); the prosecutors' October 21 filing and Meta's reply (Gazeta do Povo); the internal messages (Folha via Reuters); the runoff result (AP) and Flávio Bolsonaro's annulment plan (AP, InfoMoney).
• Uncertain: whether Meta AI's behavior came from a deliberate rule or a bug; whether the fine will be applied; and what the TSE's investigation will find now that the runoff is over.