Meta's AI Video Rules Are Still Not Working, Oversight Board Says
Meta's rules for AI-generated video remain fundamentally inadequate, the Oversight Board said in two decisions published September 17, 2026, on the rapid rise of synthetic media, meaning images, audio and video made or heavily altered with AI. Engadget
The first case centered on a realistic synthetic video of a Scottish politician. In the clip, the politician is shown saying "refugees are welcome here, even if they rape our women, because white people do that too." The video was reported to Meta but left online. Engadget
Meta told the Board it did not remove the video after the first report because it had not been flagged by trusted partner organizations, outside groups that alert the company during elections and other sensitive events, and did not appear to interfere with voting or other processes. The company also did not add an AI label because the posting account had not disclosed AI involvement, and disclosure remains the primary trigger for labeling in that workflow. Engadget
The Board said the video should have been removed under Meta's hateful conduct rules. It also said the video should have carried a prominent label stating it was AI-generated. Removal related to the hate speech rule. Labeling related to provenance, or information about how the video was made. Engadget
The second case involved AI-manipulated footage taken from a television interview with a volunteer promoting menstrual health education. The footage was altered to ridicule her. It then went viral. Engadget
A user report and a later appeal about that video were closed automatically without any AI label being applied. Meta removed the video for bullying only after the Board took up the case. Engadget
The Board recommended that Meta prioritize likely AI content for human review when it is reported. Meta has 60 days to respond to the latest recommendations. The current approach often uses a small banner that reads "AI Info." The Board called for more prominent treatment in the politician case. Engadget
The decisions continue a dispute that dates to the 2024 election cycle. In February 2024, the Board reviewed a Facebook video that falsely suggested U.S. President Joe Biden is a pedophile. It found that video did not violate Meta's rules as written. Reuters
At the time, the Board called Meta's manipulated media policies "incoherent" and insufficient and urged the company to rethink its policy on manipulated media. PBS The Board urged Meta to stop removing manipulated media that does not violate another policy and to apply a label instead. Oversight Board Meta said in April 2024 that it was changing how it handles manipulated media based on feedback from the Board and its policy review process. Meta
That change led to the current labeling system, under which Meta has begun labeling some content as "Made with AI." The Board continues to track such cases on individual case pages, including one titled "AI-Generated Video in Israel-Iran Conflict." The Board also warned in March 2026 that Meta's methods for policing AI videos are inadequate, especially at times of crisis. BBC The history goes back further. In January 2020, Facebook announced a policy to take down misleading manipulated media.
The broader context here is that detection and enforcement still depend on self-disclosure and third-party escalation. A photorealistic synthetic video can pass through user reporting, automated review, and appeal without being identified as synthetic. Flagging by trusted partners helps with election operations. It is less useful for harassment, hate, or viral ridicule.
In my view, worth noting is how labeling and removal were asked to do different jobs in these rulings, and each failed for a different reason. The politician video was checked for electoral interference while the hateful content was missed. The menstrual health video was treated as low priority until outside pressure changed its handling. Fixing that will require better provenance detection, use of classifiers to prioritize user reports about likely synthetic media, and larger, harder-to-miss labels. The long arc remains hopeful. Better origin tools and faster human review can keep creative uses of generative media while lowering the cost of abusive ones.
