Oversight Board orders Meta to remove a UK councillor deepfake

Meta's Oversight Board ruled on September 17, 2026 that a fabricated video depicting a Scottish local councillor speaking about refugees must be removed from Facebook under the company's Hateful Conduct rules, overturning Meta's decision to leave it up and finding that the company's rules on AI deepfakes are inadequate to the volume of synthetic content it now faces.

Meta had concluded, even after the board brought the case to its attention, that the video broke none of its Community Standards and did not merit an AI label, as the board's full decision states. The video was posted in November 2025 and portrayed the councillor saying: "Refugees are welcome here, even if they rape our women, because white people do that too." The board found the audio was not fully synchronised to her facial movements, indicating it was a fabrication. The board's decision states the video and the photo album it appeared in together drew more than 5,000 views before removal, and that two reports, one from the councillor herself, were never prioritised for human review.

The decision binds Meta on the individual case; the board's wider policy recommendations do not. Meta said it welcomed the ruling in both cases and would comply within seven days, and that it would act on content that is identical and in the same context where that is technically and operationally possible. The board ruled on a second post in the same sitting, a fabricated video of a Muslim campaign volunteer depicted offering health advice while exercising or eating junk food, which the Guardian reported the board found breached Meta's bullying and harassment policy.

A majority of the board went further than the removal finding. It held that even without the hate speech violation, the video met the threshold for a "High Risk AI" label, and rejected Meta's reasoning that no imminent election or crisis meant the risk of material deception was low. The board also rejected Meta's argument that the video read as satire inverting the councillor's public position, writing that there were no signals of exaggeration or parody that would make the fabrication clear. A minority disagreed, arguing that suppressing a post because it misrepresents a politician's views denies protection to a recognised genre of political speech, and that the content should carry an "AI Info" label rather than removal or demotion.

The recommendations that follow are where the practical effect sits. The board asked Meta to lower the threshold for "High Risk" labels so they are not reserved for election or crisis conditions, to add friction such as a click-through interstitial rather than a label alone, to demonetise or demote labelled content with escalating penalties for repeat sharers, and to publish annual figures on how often the labels are applied, including to content about politicians. Oversight Board co-chair Pamela San Martin said the cases showed deepfakes are increasingly used to harass and silence women in public discourse.

Councillor Dorothy McHugh, a Labour member of Dundee City Council, told The Herald on September 22 that she welcomed the removal but that the reasoning exposed a gap. "It hasn't said to remove the video because someone created a fake video of me," she said. "It's because it contains hate speech against refugees." McHugh, who is 77, said the fabrication was convincing, that it impugned her integrity, and that the process felt appalling. The board noted her seat is not contested until May 2027 and rejected Meta's view that the absence of an imminent vote reduced the case for a prompt warning.

For anyone making synthetic video of real people, the operative lines are narrow and now written down. A fabricated clip of an elected figure on a matter of public importance is label territory at minimum, satire is not assumed, and the platform's own systems decide whether a report reaches a human.

Meta has seven days from the September 17 decision to comply. The board will treat the labelling recommendations as implemented only when Meta updates its manipulated media policy, supplies evidence the penalties are being applied, and shows it has lowered the enforcement threshold for "High Risk" labels.