Entrance sign at Meta’s headquarters complex in Menlo Park, California. Credit: Nokia621 via Wikimedia Commons (CC BY-SA 4.0) Meta’s rules on AI deepfakes are “consistently and fundamentally inadequate to address the rapid rise of AI-generated content”, the Oversight Board said on 17 September. The board ordered Meta to remove two AI-generated videos from Facebook. The first ruling covers a UK local politician, and the board published the full decision.
The second ruling covers a young Muslim woman. The board overturned Meta’s original decisions in both cases, the Guardian reported. Its decisions are binding on Meta. “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, a co-chair of the board. The Scottish councillor video A Facebook user posted the video in November 2025, according to the board’s summary.
It showed the likeness of a Labour Party councillor in Scotland, making inflammatory comments about refugees that she never made. The audio was not fully synchronised with her facial movements, which indicated a fabrication, the board said. The video sat in an album with a photo of several named women, including the councillor, at an anti-far-right protest. The album and the video together drew more than 5,000 views.
Two people reported the video, including the councillor herself. Meta’s systems did not send it for human review, and both appeals failed. The other user then appealed to the board. The councillor told the board that the words “run contrary to everything [she] stands for”.
She said she had reported the video to Meta and to the police without success. What Meta and the board found Meta concluded that the video broke none of its rules and did not need an AI label, even after the board raised the case. It said no Trusted Partner organisation had flagged the video, and that it contained no banned claims about voting or the census. Meta also said the video was satirical and drew little engagement, with no election or crisis close by.
It told the board it had not penalised the user for failing to disclose AI use. A majority of the board ruled that the video broke Meta’s hateful conduct rules. It alleged “serious criminality and predatory sexual behavior” against refugees as an entire group, the majority found. A minority said the video was aimed at the politician’s alleged views, not at all or most refugees.
The majority also found that the video should have carried a “High Risk AI” label. The board did not find a breach of the bullying and harassment rules, because an elected official counts as a public figure. Six recommendations “The majority finds that Meta needs more robust policies on deepfakes,” the board said. It made six recommendations in this case.
The board asked Meta to lower its threshold for “high risk” labels beyond crises and elections. It also asked for a click-through screen, instead of a label, on content labelled as AI-generated or “High Risk AI”. Other recommendations would demote or demonetise “High Risk” content and penalise accounts that share it repeatedly. Meta should also publish yearly data on how often it applies these labels, the board said.
Meta has 60 days to respond, Engadget reported. In July, the board published a study that said top AI models were stifling political speech. The campaigner video The second ruling covers a young Muslim woman in Europe. She took part in a campaign on menstrual health education for women and girls from ethnic minority backgrounds.
AI-manipulated videos and images mocking her had tens of millions of views online, including on Meta’s platforms, the Guardian reported. The board found that one video broke Meta’s bullying and harassment policy. It showed her exercising absurdly and eating junk food. Meta automatically closed the first report on that video and the appeal.
It removed the video only after the board became involved, Engadget reported. In July, TNW reported that Meta’s AI was deleting user accounts. The board made three recommendations in this case, the Guardian reported. One would widen Meta’s definition of “unwanted manipulated imagery” to cover deepfakes of private individuals.
These show a person saying or doing things they did not say or do. Brazil’s electoral court is also defining what a deepfake is. Meta and deepfakes “These cases demonstrate a broader, troubling pattern in which women who engage publicly on issues are disproportionately subjected to harassment and misinformation,” San Martin said. Meta set up the board in 2020 to rule on content on its platforms.
The Guardian said it had approached Meta for comment. Also on 17 September, TNW reported that a German court had ruled Meta liable for scam ads on Facebook and Instagram. In July, TNW reported that the tools behind nudify apps sit on Hugging Face.
Leave a Reply