Meta’s Oversight Board orders removal of deepfake video, demands policy changes

Meta Oversight Board Demands Stronger Deepfake Protections

Meta’s Oversight Board on Tuesday overturned the company’s decision to leave up a reportedly AI-generated sexualized video on Instagram, ordering its removal and calling on Meta to strengthen protections for non-public figures targeted by deepfake intimate imagery.oversightboard

Board Rules AI Impersonation Is Non-Consensual by Default

The decision, published June 23, stems from a case in which a user reported that an AI-generated video impersonating a woman had been posted on Instagram without the depicted person’s consent, “damaging their reputation”. Meta had initially decided to leave the content up, but the Board found that the post violated the company’s Adult Sexual Exploitation policy.oversightboard

In its ruling, the Board concluded that AI-generated sexualized impersonation of real people should be treated as non-consensual by default and added to the signals Meta uses to establish lack of consent. The Board recommended Meta add a new signal to its Adult Sexual Exploitation policy specifically covering AI-generated sexualized impersonation of real people.mlex

Recommendations for Global Reporting Access

The Board issued several additional recommendations. It called on Meta to allow users to designate “connected accounts” — trusted friends, family members, or associates — who can report potential violations involving non-consensual intimate imagery on their behalf. It also recommended that Meta create a separate reporting category for AI-generated sexualized impersonation, distinct from existing categories like “harassment” or “nudity,” and make such reporting forms available globally.oversightboard

Currently, specialized reporting forms for deepfake intimate imagery are available only to users in certain jurisdictions, including Texas and Florida. The Board pushed for all users worldwide to have access to these tools, noting that non-public figures in particular “often carry the burden to self-report”.mlex

A Growing Problem Disproportionately Affecting Women

The Board also reiterated earlier recommendations that Meta replace the term “derogatory sexualized photoshop” in its policies with language referring to “non-consensual” manipulated media, and implement content credentials from the Coalition for Content Provenance and Authenticity at scale. The digital rights organization WITNESS said the decision “reflects key concerns on AI-generated sexualized impersonation,” affirming that consent should be placed at the center of content moderation in this area.witness

The ruling follows the Board’s February announcement that it would take up the case, and builds on a July 2024 decision in which the Board said Meta’s deepfake policies needed updating. The Board has warned that advances in AI tools have led to a proliferation of non-consensual sexualized content that disproportionately harms women and girls.kark