Particle.news

Oversight Board Orders Meta to Remove Two Harmful Deepfakes

The board found Meta’s moderation, labeling and prioritization failed to stop AI-manipulated videos that harass private women or spread hate, recommending specific policy and enforcement changes.

Overview

  • The Oversight Board on Thursday, September 17, 2026, overturned Meta’s prior choices and required removal of an AI-generated video that bullied a private Muslim woman and a deepfake portraying a Scottish councillor making hateful comments about refugees.
  • The Board found Meta’s rules and systems did not treat falsified speech or actions as covered harms, and that automated triage left both posts off the human-review queue and unlabeled as AI-generated.
  • It recommended broad policy changes including expanding the definition of “unwanted manipulated imagery” to cover falsified words and actions, adding violating items to Meta’s Media Matching Bank, and using AI-detection signals to raise report priority.
  • The Board insisted on stronger labeling and enforcement: lower thresholds for ‘High Risk’ AI labels, add viewing friction like interstitials, demonetize or demote labeled content, and impose escalating penalties for repeat sharers.
  • Decisions include minority views about political-speech trade-offs, and the Board’s binding removals will take effect only if Meta implements its recommendations and follows the required removals.