Meta Oversight Board Orders Removal of AI Deepfakes, Calls Company's Safeguards Inadequate

Meta Oversight Board Condemns Inadequate Safeguards for AI Deepfakes

The Meta Oversight Board issued a scathing rebuke of the company's approach to AI-generated content, declaring Meta's deepfake rules "consistently and fundamentally inadequate to address the rapid rise of AI-generated content" on its platforms1

1

2

2

. The quasi-independent body ordered Meta to remove deepfake videos targeting two women and issued nine policy recommendations across both cases2

2

. Meta has 60 days to respond to these recommendations, though individual content decisions are binding on the company3

3

.

Scottish Politician Targeted by AI-Generated Deepfake Video

A Facebook user posted an AI-generated deepfake video in November 2025 showing a Labour Party councillor in Scotland making inflammatory comments about refugees that she never made2

2

. The fabricated video depicted the councillor saying "refugees are welcome here, even if they rape our women, because white people do that too"4

4

. The audio was not fully synchronized with her facial movements, indicating fabrication2

2

. The video sat in an album with photos from an anti-far-right protest and drew more than 5,000 views2

2

. The councillor told the board the words "run contrary to everything [she] stands for" and described the effect as "quite traumatic"1

1

.

Meta's Failed Response to Reported Content

Meta told the board it didn't remove the deepfake video when reported because no "trusted partner" organizations had flagged it and it didn't appear to interfere with voting or other processes1

1

. The video was not labeled as AI-generated because the person who posted it didn't choose to disclose it was made with AI1

1

. Meta's systems did not send the video for human review, and both appeals from the councillor and another user failed2

2

. Even when the board raised the case directly, Meta concluded the video broke none of its rules and did not need an AI label2

2

.

Harassment of Women Through AI-Generated Content

The second case involved a young Muslim woman in Europe who participated in a campaign on menstrual health education for women and girls from ethnic minority backgrounds2

2

. AI-manipulated videos and images mocking the Muslim campaigner had tens of millions of views online, including on Meta's platforms4

4

. One video showed her exercising absurdly and eating junk food, falsely depicting her consuming junk food while giving advice about mindful eating5

5

. Meta automatically closed the first report and appeal, removing the video only after the Meta Oversight Board became involved1

1

.

Board Finds Multiple Policy Violations

A majority of the board ruled that the Scottish politician video broke Meta's hateful conduct rules by alleging "serious criminality and predatory sexual behavior" against refugees as an entire group2

2

. The board determined the video should have received a "High Risk AI" label and been prominently labeled as AI-generated content1

1

. In the second case, the board found the video of the Muslim campaigner violated Meta's bullying and harassment policy4

4

. "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, Meta Oversight Board co-chair1

1

.

Nine Policy Recommendations for Stronger Content Moderation

The board made six recommendations in the Scottish politician case and three in the Muslim campaigner case2

2

. Key recommendations include lowering the threshold for high-risk AI content labels beyond crises and elections, and replacing labels with click-through warning screens that require users to click before viewing certain AI-generated content2

2

4

4

. The board called for algorithmic demotion of content labeled as high-risk AI content, making such posts less prominent in users' feeds4

4

. Other measures include stricter penalties for accounts that repeatedly share high-risk AI content and increased transparency through yearly published data on how often Meta applies AI labels2

2

. The board also recommended widening Meta's definition of "unwanted manipulated imagery" to cover AI deepfakes of private individuals saying or doing things they did not say or do2

2

.

Meta's Limited Response to Previous Recommendations

Meta's Oversight Board has repeatedly criticized the company for its approach to AI-generated content and urged new rules, but Meta has made few changes to its AI labels, which typically consist of a small banner saying "AI Info"1

1

. The company hasn't agreed to implement many of the board's recommendations on AI labeling in other recent cases1

1

. Meta welcomed the board's decision in both cases and said it will comply within seven days, though the company's broader policy recommendations are not binding5

5

. The board's criticism raises questions about whether Meta genuinely prioritizes addressing misinformation and harassment through AI deepfakes on Facebook and other platforms.