Full report here.
São Paulo, 17 September 2026 - Meta is approving AI-generated ads that appear to break Brazil’s strict election rules, just weeks away from voters going to the polls to elect a new President.
Fifteen ads were submitted, with 13 being approved by Meta’s AI moderation system, including one inciting violence against a lawmaker and another calling for supporters to ‘retake Brasilia’ in a repeat of the January 8th attempted coup. Other approved ads targeted lawmakers with hate speech, falsely celebrated Trump’s official endorsement of Flávio Bolsonaro, and pushed the conspiracy theory that Jair Bolsonaro is a political prisoner. All of the ads appear to breach Meta's own community standards and Brazil's strict Superior Electoral Court (TSE) rules. The research was conducted by corporate accountability group Ekō in August.
Researchers submitted the ads on Meta, each paired with an AI-generated image, modeled on popular conspiracy theories and narratives being promoted by Brazil’s far-right. The images were generated using mainstream AI chatbots, including ChatGPT, Grok, and Gemini. The two ads Meta rejected were flagged only for "deceptive or misleading business practices" – not for hate speech, incitement, or election disinformation.
Vicky Wyatt, Ekō Campaign Director said: “Meta learned nothing from its role in fuelling Bolsonaro's attempted coup in 2022. Four years on, its systems are still waving through ads inciting violence, hate speech, and disinformation — weeks before Brazilians go to the polls again.
“This research shows just how badly Meta's systems are failing. Meta has pulled back on safety at the exact moment AI tools have made it easier than ever to generate hyperrealistic deepfakes in minutes. This is the business model of disinformation and hate. Meta profits from political advertising whether it's true or not — and profit is the only metric that's ever mattered to this company.”
Brazil has some of the world's strictest AI election rules – requiring all AI-generated electoral content to be visibly labeled, banning synthetic content that could harm electoral integrity, and prohibiting foreign-sourced electoral advertising. Yet Ekō researchers found that AI-generated ads depicting real candidates and Supreme Court justices saying things they never said were submitted to Meta from a US-based account targeting Brazilian voters, with no AI disclosure labels. This appears to violate multiple TSE rules at once, including restrictions on foreign electoral donations and foreign ad placements.
The ads also clearly pushed the limits of discriminatory content and hate speech. One ad contained dehumanizing language to attack transgender members of Congress – a form of political gender-based violence that Brazilian law treats as a criminal offense in its own right.
Beyond potentially breaking Brazilian law, the approved ads violate Meta's own policies on multiple fronts – breaking its rules on hateful conduct, violence and incitement, misinformation, and political ad transparency. None of the ads carried Meta's required "Social Issues, Elections or Politics" disclaimer, and none were flagged as election-related content despite directly referencing named candidates and the vote itself. The same failure has been documented in Ekō investigations in India and Germany, where Meta repeatedly approved AI-generated ads inciting violence, spreading hate speech, and pushing election disinformation.
All ads were submitted for review through Meta's standard ad approval process without being published or shown to any Meta users; all ads were removed by researchers immediately upon approval or rejection, meaning that they were never seen by Meta users.
This is taking place as Meta continues to roll back the systems built to catch this kind of content. In January 2025, Meta ended its third-party fact-checking program, announced it would roll back moderation on "hot-button" topics, and made massive cuts to its Trust and Safety teams. Disinformation researchers warned that these moves would cause a surge in false narratives. Brazil is particularly exposed as its TSE rules rely heavily on Meta's own systems and good-faith compliance, leaving a gap that this investigation shows Meta is not filling.
ENDS
Ad text and images used in this investigation can be shared with media on request for corroboration purposes, but must not be published.
For more information about a current Ekō campaign or to arrange an interview with a spokesperson please contact us at
