Meta Must Do More Against Harmful Deepfakes Containing Hate Speech
17 septembre 2026
The Oversight Board has found that a deepfake video of a UK local politician should be removed from Facebook because it contains hate speech against refugees, overturning Meta’s decision to leave it up. The Board found Meta needs to do much more to effectively address such content, including by increasing penalties for sharing harmful deepfakes, to protect users from potential deception.
Why This Matters
The case surfaces monumental challenges facing societies that social media companies must address – the increasing realism of deepfakes depicting elected officials misrepresenting their views on critical issues and misleading the public, and misinformation that fuels hatred against minorities, including immigrants. Women politicians, journalists and activists are disproportionately targeted by deepfakes that misrepresent their views, including as part of harassment campaigns, and less prominent public figures such as local politicians and activists often lack the resources to respond effectively. The spread of deepfakes erodes trust in all public information and challenges the boundaries between what is protected political speech and what violates Meta’s rules on misinformation, hate speech or bullying and harassment.
About the Case
In November 2025, a Facebook user posted a short video showing the likeness of a Labour Party councillor in Scotland. The video, which appears to be AI-generated, portrays her saying: “Refugees are welcome here, even if they rape our women, because white people do that too.”
The portrayal is realistic, though on closer inspection appears synthetic: the audio is not fully synchronized to her facial movements, indicating that it is a fabrication. The video was part of an album containing a photograph of several named women at an anti-far-right protest, including the councillor. The album and video together received more than 5,000 views, over 50 comments, over 20 reactions and more than 10 shares.
Two individuals, including the depicted politician, reported the video for violating the Bullying and Harassment policy, but Meta’s systems did not prioritize the post for human review, and it remained on Facebook. Both users appealed to Meta, but the post was kept up and not reviewed by a human. One of the users, not the politician involved, then appealed to the Board. When the Board brought the case to Meta’s attention, the company concluded the video did not violate its Community Standards and that it did not merit an AI label.
Key Findings
The Board evaluated this case with reference to three of Meta’s policies – Hateful Conduct, Misinformation, and Bullying and Harassment. The Board finds that Meta should have removed the AI-generated video under its Hateful Conduct policy.
The majority of the Board finds that the video violated Meta’s Hateful Conduct rules specifically because it alleges serious criminality and predatory sexual behavior against refugees as an entire group, and not against some refugees as Meta claimed.
In ruling on the case, the Board also considered whether the video, had it not violated the Hateful Conduct rules, would have met the threshold for removal under the misinformation policy. A majority of the Board finds that removal would not have been warranted for misinformation, but that it should have received a “High Risk AI” label, and in its analysis concludes that Meta’s rules on AI labeling of deepfakes are inadequate.
The majority finds that Meta needs more robust policies on deepfakes, including expanding the situations when “high risk” labels can be applied, more measures to reduce the spread of deceptive AI content, increasing the penalties for accounts that repeatedly share it, and more transparency on data around when AI labels are applied.
A minority of the Board disagrees that the video should be considered hate speech because it is directed at the alleged views of the politician, rather than at all or most refugees; and that because the video is political speech, the threshold for removal should be higher.
There were also minority views on the Board’s findings under the misinformation rules. One minority said that a “High Risk” label is too restrictive and that a lesser “AI Info” label should have been applied. Another, separate minority said that the gravity of harms, coupled with the fact that Meta does not in practice apply many AI labels, would justify removal.
The Board finds that the video would not require removal under the Bullying and Harassment rules because the elected official is a public figure and therefore is entitled to less protection than private individuals under the policy.
The Oversight Board’s Decision
The Board overturns Meta's decision to leave up the content, requiring it to be removed under the Hateful Conduct Community Standard. In considering this case, the Board also finds shortcomings in labeling policies under the Misinformation Community Standard and makes the following recommendations regarding that policy.
The Board recommends that Meta:
- Change its manipulated media policy to lower the threshold for "high risk” labels, making sure this is not a crisis or election integrity measure only.
- Increase friction for users viewing content labeled as AI-generated or “High Risk AI”, such as using an interstitial (a screen requiring click-through to view content) instead of a label.
- Subject all “High Risk” labeled content to demonetization, demotion or removal from recommendations, with escalating penalties for accounts that repeatedly share "High Risk" labeled content.
- Ensure that the rules and examples related to voting or census interference are phrased as globally applicable and not limited to the U.S. context.
- Publicly disclose annual data in its Transparency Center on the number of times it applies the “High Risk” and “High Risk AI” labels to content, including the number of times these labels are applied to content from or about politicians.
- Ensure that all labels under the Misinformation policy (“AI Info,” “High Risk” or “High Risk AI”) can be seen on content in the Meta Content Library, and that content that has received these labels is searchable via a filter.
The Board also reiterates the importance of its previous recommendation in the AI Generated Video in Israel-Iran Conflict decision, calling for Meta to publish a clear explanation of penalties for failure to self-disclose digitally created or altered content. It should provide criteria for penalties and list which account features are consequently limited and for how long.
Futher Information
To read public comments for this case, click here.
