Brazil's TSE Faces 768 Unlabeled AI Election Posts

Brazilian fact-checking agency Lupa counted 1,423 AI-made or AI-altered election posts from 16 August to 22 October. Of those, 768—54%—carried no AI label, despite electoral court rules requiring one. Candidates, ordinary users and anonymous accounts all helped fill social media feeds during the campaign season, while the court’s enforcement process moved through formal channels.

Lupa and the Recod.AI laboratory at the University of Campinas (Unicamp) run the live VigIA dashboard. It was updated Friday, 23 October, before Brazil’s presidential runoff on Sunday, 25 October. The tracker counted 1,486 posts in total: 1,423 confirmed AI content, 32 with no AI and 31 that remained inconclusive. Lupa may change those totals as it reviews posts.

Who Put the Posts Online

Ordinary users made 560 of the confirmed posts, or 39.4%. Official candidate accounts made 548, or 38.5%; anonymous, unknown or avatar accounts made 300, or 21.1%; and political parties made eight. Among the unlabeled material, 386 posts came from ordinary users, 188 from anonymous accounts, 185 from candidate accounts and five from parties. Missing labels weren’t confined to anonymous corners of the internet.

Lupa founder Cristina Tardáguila said campaigns know they can face punishment, so they leave misuse to militants. Researcher Beatriz Farrugia told Agência Brasil that courts would need platforms’ help to identify anonymous accounts. Those accounts made up a substantial share of unlabeled posts, but identifiable candidate accounts also published them.

The confirmed material included 707 synthetic posts, 520 deepfakes and 196 deepfakes mixed with other synthetic content. A deepfake is AI-made or AI-altered media that shows a real person doing or saying something they never did. Images accounted for 751 posts; videos accounted for 667.

Instagram carried posts 794 times and Facebook 690; X had 105, TikTok 72, WhatsApp three and YouTube three. A single post could appear on more than one platform. By purpose, 768 posts promoted a candidacy, 360 attacked someone, 264 were humor or satire, and 31 contained neutral information.

President Luiz Inácio Lula da Silva, of the left-wing Workers’ Party (PT), appeared in 509 posts, more than any other public figure. Flávio Bolsonaro, a conservative senator of the Liberal Party (PL), appeared in 247, while his father, former president Jair Bolsonaro, appeared in 120. Supreme Court justice Alexandre de Moraes appeared in 56, jailed ex-banker Daniel Vorcaro in 51 and congressman Nikolas Ferreira in 48.

Rules, Platforms and the Enforcement Pipeline

The TSE requires an explicit, prominent label on AI campaign content. Article 9-B of Resolution 23.610, amended by Resolution 23.755 of 2 March 2026, also bans deepfakes that favor or harm a candidacy. The rules set fines of BRL 5,000 to BRL 30,000, require platforms to remove such content, and let the court cancel a candidacy or mandate for abuse of power.

A separate restriction bans new AI content that alters a candidate’s voice or image from 72 hours before a vote until 24 hours after it. Before the first round, Lupa found 64 such posts between 1 and 5 October; candidates made only four. The exact hour the restriction took effect before the runoff wasn’t stated.

On 15 October, Lupa reported that candidates and parties had 37 posts flagged in the first 11 days of the campaign. Twenty-nine remained online on Tuesday, 13 October; authorities had removed the others. Lupa sent the cases to the TSE on 11 September, and the TSE forwarded them to the electoral prosecutor-general’s office on 16 September. The TSE told Lupa complaints should go directly to that office.

Meta said it was committed to protecting the electoral process and that Lupa’s sample didn’t reflect the scale of its work. TikTok said it removed 348,963 videos in Brazil between 1 June and 6 September for civic-integrity, election, misinformation and deceptive-AI violations. X didn’t respond to Lupa before the first-round report.

The tracker doesn’t establish that any post decided the first round, and no one has measured the posts’ effect on votes. Lupa said it would keep monitoring AI content through the runoff, with its final VigIA report listed for November. The count records a flood of synthetic campaign material—and the formal systems tasked with labeling, identifying and removing it.