Meta's own oversight board orders takedown of two UK deepfake videos
Meta's quasi-independent oversight board ruled the company should have removed two AI-generated videos — one smearing a Scottish councillor, one mocking a Muslim campaigner — and called Meta's deepfake safeguards "fundamentally inadequate."
What happened: Meta's oversight board, the body set up in 2020 to referee tough content calls, ruled that Facebook wrongly left up two AI-generated deepfake videos. One falsely showed a Scottish Labour councillor making inflammatory, false claims that refugees commit sexual assault. The other mocked a young Muslim campaign volunteer working on menstrual health education, showing her doing absurd exercises and eating junk food; related clips racked up tens of millions of views. The board ordered both removed and said Meta's safeguards against deepfakes are "consistently and fundamentally inadequate."
Why it matters: This is a rare case of a body with binding authority over Meta publicly stating the company's AI-fakery defenses are failing, even when problems are flagged directly. Board co-chair Pamela San Martin said deepfakes are increasingly used to harass and silence women in public life, and that both cases fit a pattern of women engaging publicly being singled out for fabricated content. For a Scottish local politician and a health campaigner, the practical harm was reputational damage and abuse that spread widely before any takedown happened.
How it works, plainly: The board said the councillor clip showed signs of being AI-made, including audio not fully matching lip movements, and it should have triggered removal under Meta's hateful conduct rules for smearing refugees as a group, plus a "high risk AI" label. Yet when the board itself raised the video with Meta, the company decided it broke no rules and didn't need labeling. The Muslim volunteer video was ruled a breach of bullying and harassment policy instead, since it targeted an identifiable private individual with fabricated behavior she never engaged in.
The rollout: Beyond ordering the two takedowns, the board recommended nine policy changes: expanding when "high risk" AI labels apply, demoting such labeled content in feeds, adding click-through warning screens before users view flagged AI content, tougher penalties for repeat sharers, more transparency on labeling data, and a broader definition of "unwanted manipulated imagery" covering fake depictions of private individuals. These are recommendations, not binding orders, and Meta has not yet said whether it will adopt them. Meta has been approached for comment.