Meta Ordered to Remove Deepfakes as Board Finds AI Safeguards Inadequate
Meta's Oversight Board has ordered Meta to remove two deepfake videos from Facebook and described the company's protections against AI-generated content as "consistently and fundamentally inadequate." The Guardian
One video targeted a Labour Party councillor in Scotland. The other targeted a young Muslim woman. Both rulings are binding on Meta.
What the videos showed
The first clip falsely showed the councillor saying: "Refugees are welcome here, even if they rape our women, because white people do that too." The Board judged the clip apparently AI-generated because the audio did not fully line up with the councillor's facial movements. A deepfake works like a synthetic impersonation, built to look and sound like a real person.
Meta was told about that assessment. Even when the Board raised the case directly, Meta decided the video violated no policy and needed no AI label. The Board disagreed on both points.
It said the post should have been removed under Meta's hateful conduct rules because it accused refugees as a group of criminal and predatory sexual behaviour. It also said the video should have carried a "high risk AI" label.
The second ruling involved a young Muslim woman in Europe who volunteers in a campaign to improve menstrual health education and reduce stigma for women and girls from ethnic minority backgrounds. The fake video showed her giving health advice while exercising in absurd ways or eating junk food. Related manipulated videos and images mocking her drew tens of millions of views online, including on Meta platforms. The Board found that video violated Meta's bullying and harassment policy.
What the board wants changed
Across the two decisions, the Board made nine policy recommendations. They include having algorithms demote content labelled "high risk" so it appears less often in feeds. They also include making AI content harder to view, for example behind a warning screen that requires a click to proceed. The Board also urged Meta to widen its definition of "unwanted manipulated imagery" to cover deepfakes showing a private individual saying or doing things they never said or did.
The Board is a quasi-independent body set up by Meta in 2020 to rule on content on its platforms. It had earlier warned that Meta's methods for policing AI video are inadequate, especially at times of crisis. BBC
The broader context here is enforcement, not just rulemaking. The Board did not fault Meta for lacking written bans on hateful conduct or bullying and harassment. It faulted detection, labelling and removal in practice, including internal review that left one video up after external scrutiny. For policymakers and platform specialists, that distinction matters. A demotion and click-through system would move from simply disclosing AI content toward limiting how widely it spreads, which raises practical questions about detection thresholds, appeal rights, and how review queues work in a crisis. It also tests whether rulings on single posts can drive wider reform.
