Meta Ordered to Remove AI Deepfake After Policy Failure

Meta's Oversight Board has reversed the company's decision to leave up an AI-generated video that impersonated a woman on Instagram, ruling that the content violated Meta's policies against non-consensual intimate imagery and exposing significant gaps in how the platform handles deepfakes targeting private individuals.

The case centered on an eight-second video posted in September 2025 showing a woman adjusting her dress with underwear visible in several frames. According to the Oversight Board, the video was AI-generated and depicted someone without their consent—the person had already closed her Instagram account by the time a friend appealed the content to the Board.

Meta's automated detection system flagged the post as potentially harmful content with high virality risk but never prioritized it for human review. Two users reported the video, and one appealed Meta's decision to leave it up, yet the content remained visible. Only when the Oversight Board escalated the case did Meta's subject matter experts review it, ultimately restricting visibility to adults rather than removing it entirely.

The consent problem

The Board determined the video violated Meta's Adult Sexual Exploitation Policy, which prohibits non-consensual intimate imagery. The ruling hinged on three criteria: the content appeared non-commercial and set in a private context, the woman was "near nude," and consent was absent.

Meta defended its original decision by stating it had "no indication that the individual depicted in the video was 'a real person'" because the person depicted never reported the content themselves. The company currently relies on signals including direct reports from those depicted, captions suggesting revenge context, or reports from law enforcement and media organizations.

The Oversight Board rejected this framework for AI-generated content. It ruled that AI-generated impersonation should be considered non-consensual by default and added as a signal for lack of consent in Meta's policies.

Why it matters

This decision arrives as governments worldwide establish new regulations for AI-generated content. The European Union has agreed to prohibit AI systems that create non-consensual sexually explicit content, while India, the UK, and Spain are implementing their own platform rules. The Board noted that sexualized deepfakes cause reputational and psychological harm that disproportionately affects women and girls, creating a chilling effect on social and political participation.

The ruling particularly benefits non-public figures who lack the resources or visibility to quickly address abuse. Experts consulted by the Board emphasized that victims benefit most from rapid removal and reporting mechanisms that don't require self-reporting—a process that can compound trauma.

Recommended changes

The Oversight Board issued several recommendations to Meta. The company should add AI-generated sexualized impersonation as an explicit signal of non-consent in its Adult Sexual Exploitation policy. Meta should also allow users to designate "connected accounts"—trusted friends or family members—who can report violations on their behalf.

Additionally, the Board called for AI-generated sexualized impersonation to be listed as a distinct reporting category, separate from general harassment or nudity complaints, and made available globally. The Board reiterated earlier recommendations to implement content credentials at scale following standards from the Coalition for Content Provenance and Authenticity.

The decision and recommendations were first reported by the Oversight Board.