Meta approved 53% of election disinformation ads in Brazil experiment

The failure of Meta's automated systems to filter blatant disinformation suggests that AI-driven moderation remains insufficient for high-stakes political events. For the streaming and social video ecosystem, this highlights a critical vulnerability where microtargeting tools can be exploited by foreign actors despite existing regulatory frameworks like Brazil's Superior Electoral Court resolutions. As platforms increasingly rely on algorithmic enforcement to manage massive ad volumes, the inability to distinguish between political discourse and prohibited misinformation creates significant legal and reputational risks. Industry observers should monitor whether the Brazilian government imposes fines or suspensions on Meta similar to the 2024 restrictions placed on X.

Meta's AI-driven ad moderation systems have faced mounting criticism from civil society organizations and regulators across Latin America. In July 2025, Brazil's Superior Electoral Court approved new rules requiring platforms to label AI-generated political content, mandating that any synthetic media used in campaign materials be clearly identified to voters. The TSE also established that platforms must remove content flagged as disinformation within 24 hours of a judicial order, creating a compliance framework that directly tests whether Meta's automated systems can meet legal deadlines. Amnesty International's experiment, which found that eight of 15 mock disinformation ads were approved, suggests Meta's current filtering infrastructure may struggle to satisfy those obligations at scale during the 2026 presidential election cycle.

The regulatory pressure on Meta extends beyond Brazil's electoral court. In August 2025, Brazil's data protection authority ANPD opened a formal investigation into Meta's use of personal data for AI training, adding a separate compliance burden that could result in fines of up to 2% of Meta's Brazil revenue. Meanwhile, the European Union's Digital Services Act has already forced Meta to publish transparency reports on political ad spending, and the European Commission issued a preliminary finding in April 2025 that Meta's ad library failed to meet DSA transparency requirements, specifically citing gaps in advertiser identity verification and incomplete disclosure of targeting parameters. These parallel enforcement actions across jurisdictions indicate that Meta's ad integrity failures are not isolated to any single market but reflect systemic weaknesses in its automated moderation pipeline.

On the technical side, Meta has acknowledged the limitations of its AI moderation tools while defending their overall effectiveness. In a March 2025 blog post, Meta reported that its AI systems now detect and remove over 98% of violating content before users report it, though the company noted that political disinformation remains among the hardest categories due to the contextual judgment required to distinguish satire, opinion, and factual claims. Independent researchers have challenged those figures. A study published by the NYU Center for Social Media and Politics in February 2025 found that Meta's ad transparency tools failed to flag approximately 40% of political ads containing misleading claims in the United States, suggesting the problem is not limited to Portuguese-language content or emerging markets. For streaming and video platforms that increasingly carry political advertising, the findings underscore the difficulty of relying solely on algorithmic enforcement for content that requires nuanced human judgment.