Meta Oversight Board orders deepfake removal, calls AI rules inadequate

Meta’s Oversight Board has ruled that the company must take down an AI-generated video of a Scottish Labour councillor that falsely showed her endorsing sexual violence by refugees. 

The company initially decided to leave it up, and beyond overturning that decision, the oversight board has also warned that Meta’s handling of deepfakes puts users at risk of deception.

What fabricated video must Meta take down? 

In November 2025, a clip using the likeness of a local Labour Party politician in Scotland surfaced on Facebook. The video portrayed the likeness saying, “Refugees are welcome here, even if they rape our women, because white people do that too.”

Meta’s Oversight Board described the portrayal as realistic but synthetic, noting the audio did not line up with her facial movements. The video was found in an album which also contained a photograph of several named women, including the councillor, at an anti-far-right protest.

Two people, one of them the councillor herself, flagged the video under Meta’s bullying and harassment rules, but Meta’s systems never routed it to a human reviewer, and it stayed up through two appeals. 

A third user then took the case to the Board, and even after they flagged it, Meta maintained that the video broke no rules and did not warrant an AI label. The councillor told the Board the video was “quite traumatic.” 

A majority of the Board found the clip violated Meta’s Hateful Conduct policy because it accused refugees as a whole group of serious criminality and predatory sexual behavior, rejecting Meta’s claim that it targeted only some refugees. The Board did not treat it as removable misinformation, but said it should have carried a “High Risk AI” label.

The Board did agree on one point in Meta’s favor, which is that the bullying and harassment rules did not require removal because an elected official is a public figure entitled to less protection than a private person.

What recommendations does the Oversight Board have for Meta?

The board also stepped in regarding a second case, which involved an AI-manipulated interview of a woman promoting menstrual health education. Following the board’s intervention, Meta pulled the video for bullying. 

The two rulings prompted the Board to call Meta’s safeguards “consistently and fundamentally inadequate.”

The Board recommended that Meta lower the threshold for “high risk” labels so they are not reserved for elections or crises and replace some labels with a click-through interstitial. It also suggested that high-risk AI content should be demonetized, demoted, or removed from recommendations, with escalating penalties for repeat sharers. 

It also wants Meta to prioritize suspected AI content for human review when reported, publish annual data on how often it applies AI labels, including to posts about politicians, and phrase its voting and census rules to apply globally rather than only in the United States. Meta has 60 days to respond to the board. 

Governments and platforms globally are tightening their rules on synthetic media. For instance, China’s Supreme People’s Court set liability rules on September 7, which state that creating or sharing an AI replica of someone’s face or voice without permission is an infringement. Cryptopolitan reported that Beijing separately removed more than 14,000 AI products in the first phase of its “Qinglang” cleanup earlier this year. 

The European Union’s Code of Practice on Transparency of AI-generated Content pushes providers to mark AI output and deployers to label deepfakes. 

Google Play’s AI-Generated Content policy prohibits users from posting non-consensual deepfake sexual material and election content that is “demonstrably deceptive or false.”

Only a few users are allowed to use AI in writing undisclosed on Medium, and the platform forbids AI-generated disinformation outright. 

If you're reading this, you’re already ahead. Stay there with our newsletter.