Meta Oversight Board Orders Removal of AI Deepfake Videos, Proposes Nine Policy Changes

Summary

Meta’s Oversight Board issued rulings this week ordering Facebook to remove two AI-generated videos that the platform had declined to take down under existing rules. One video falsely depicted a Scottish Labor councilor making inflammatory comments about refugees. The other manipulated the image of a young Muslim campaign volunteer. The board ruled both violated Meta’s existing policies on hateful conduct and harassment, and said viewers should have seen stronger AI-disclosure warnings before watching. Alongside the removal orders, the board proposed nine policy changes, including reduced algorithmic distribution for high-risk AI content, warning screens before users see certain manipulated media, and stricter penalties for accounts that repeatedly distribute deceptive deepfakes.

Why it matters

The Oversight Board cannot force Meta to adopt its policy recommendations, but the rulings put specific enforcement failures on the record. For families trying to understand what safeguards exist on platforms they use, the proposed changes, including slowing the algorithm on content flagged as high-risk AI manipulation before moderation can review it, are more substantive than a label. Whether Meta implements them is the next question.

HTD Says

Meta’s own Oversight Board ruled that the platform failed to remove AI deepfake videos targeting real people, and the nine-point policy overhaul it proposed goes well beyond slapping a label on manipulated content. This is an interesting first-step development in deepfake mitigation. I don’t expect it to go very far, but the encouraging sign is that it is being mentioned publicly.

I’d like to see these nine recommended points, as well as which, if any, are actually implemented. This will be extremely difficult to govern and police. Meta isn’t the only platform in question. Hundreds or thousands of media-sharing services exist, and this would probably become a whack-a-mole effort.

Identification is one thing. Enforcement is a completely different Herculean effort.

Source:

The Guardian