The Oversight Board has directed Meta to remove AI-generated deepfakes targeting women and issued nine recommendations to bolster the company's response to manipulated media. Cases reviewed include deepfakes of a UK local politician and an AI video of a female campaigner; the board called Meta's safeguards 'consistently inadequate'. Key proposals include automated removals, clearer AI labelling, prioritised human review for harassment, limits on monetisation and distribution of high-risk content, and global voting protections.
Oversight Board Urges Meta To Remove AI Deepfakes Targeting Women, Proposes Nine Policy Reforms
Phone with Meta on screen, Facebook logo ©Adobe Stock Images
The independent Oversight Board has instructed Meta Platforms to remove AI-generated deepfakes that target women and has issued nine policy recommendations to strengthen the company's handling of manipulated media across its services.
Board Findings and Notable Cases
In a statement dated 17 September 2026, the board reviewed cases that included deepfake videos of a UK local politician flagged for hate speech and an AI-generated video depicting a female campaigner. The board said Meta's existing safeguards have been 'consistently inadequate' in responding to the accelerating volume and sophistication of AI-manipulated content.
Key Recommendations
- Automated Removal: Introduce automated processes to remove deepfakes that meet high-risk or clear-harm criteria while ensuring due process and appeal pathways where appropriate.
- Clear AI Labelling: Improve and standardise labels for AI-generated content, and publish how often those labels are applied.
- Prioritised Human Review: Require prompt human review for reports of deepfake harassment and use AI-generated signals to surface priority cases for moderators.
- Limit Monetisation and Distribution: For high-risk AI content, restrict monetisation and reduce or remove distribution via recommendation systems.
- Escalating Penalties: Impose stronger penalties on accounts that repeatedly upload harmful manipulated media.
- Expand Policy Scope: Extend the 'unwanted manipulated imagery' policy to explicitly cover deepfakes that depict private individuals.
- Global Voting Protections: Apply rules protecting against voting and census interference globally, not just in the US.
- Transparency Reporting: Publish annual figures in Meta's Transparency Centre showing the frequency and enforcement of AI-generated content labels.
- Use AI Signals Responsibly: Leverage AI detection signals to help identify problematic content while safeguarding against false positives.
Context and Next Steps
The board emphasised that its removal decisions and its broader policy recommendations are distinct: the latter propose changes to Meta's rules and enforcement systems rather than asserting those measures are already implemented. The recommendations aim to broaden Meta's content-moderation framework as generative AI makes manipulated media easier to produce and distribute.
Board assessment: Meta's current safeguards are 'consistently inadequate' in the face of growing AI-manipulated content.
The board also briefly referenced Meta's stock price in its statement, noting the commercial and reputational stakes for the company as it updates its policies. Meta will now consider the removal rulings and weigh the nine recommendations as it adapts its systems and policies.
Help us improve.
