A fabricated video of a Scottish councillor was flagged to Facebook and Meta left it up. It carried no AI label either. On September 17, 2026, Meta’s Oversight Board overturned that call and a second one. It ordered two AI-generated videos taken off Facebook and called the company’s deepfake safeguards “consistently and fundamentally inadequate” for the rapid growth of synthetic content on its platforms.
The councillor case
The first video, posted in November 2025, showed a Labour Party councillor in Scotland apparently saying: “Refugees are welcome here, even if they r*** our women, because white people do that too.” The councillor never said it. The board concluded the clip looked synthetic because the audio did not line up with the councillor’s facial movements. The councillor called hearing a replicated version of their own voice “quite traumatic.”
Two people reported the video, the councillor among them. According to The Next Web, Meta’s systems never sent it for human review, and both appeals failed before the case went to the board. Meta has said the post had not been flagged by any of its “trusted partner” organizations and did not appear to interfere with voting or other electoral processes, so it got no AI label.
The board saw it differently. It said the video broke Meta’s hateful conduct rules because it attributed criminal and predatory behavior to refugees as a group, so Meta should have removed it. At the very least, it should have carried a prominent AI-generated label.
The campaigner case
The second case involved a young Muslim woman who campaigns on menstrual health education for girls from ethnic minority backgrounds. AI-manipulated videos and images mocking her were viewed tens of millions of times. The Next Web reports that Meta closed the first report and appeal automatically and removed the content only after the board took up the case. The board found that the video violated Meta’s bullying and harassment policy.
Nine recommendations
Across the two cases, the board made nine policy recommendations. The main ones:
- apply “high risk” AI labels in more situations, not just around elections and crises
- have algorithms reduce the distribution of content classified as high risk and show it less prominently in feeds
- add warning screens that users must click through before viewing certain AI-generated content
- increase penalties for accounts that repeatedly share deceptive AI content
- publish more data on when AI labels are applied
“From politicians to private citizens, AI-generated deepfakes are increasingly being used to harass and silence women from engaging in public discourse,” said Pamela San Martin, an Oversight Board co-chair. She added that the cases show “a broader, troubling pattern” in which women who speak publicly on issues face a disproportionate share of harassment and misinformation.
What happens next
Meta set up the Oversight Board in 2020 to review difficult moderation decisions. Its rulings on individual posts are binding, so both videos have to come down. Its policy recommendations are not binding. Meta has 60 days to respond to them, and that response will show how seriously it takes the board’s view that labeling alone can no longer keep up with deepfakes.
