Oversight Board calls Meta's AI deepfake rules inadequate

Meta's Oversight Board overturned the company in two decisions published on September 17, 2026, ordering Meta to remove an AI-generated video of a Scottish politician and a manipulated TV interview of a health campaign volunteer.

The Board said Meta's safeguards are "consistently and fundamentally inadequate to address the rapid rise of AI-generated content on its platforms," according to Engadget's account of its statement. Both cases were user appeals, and both decisions are binding on Meta once the company implements them.

In the first case, 2026-034-FB-UA, a Facebook user posted an album in November 2025 containing a short video that used the likeness of a Labour Party councillor in Scotland. The Board's decision describes the video as a realistic but synthetic portrayal in which the audio is not fully synchronized to her facial movements. It shows her appearing to say, "Refugees are welcome here, even if they rape our women, because white people do that too." The councillor did not say it. In a statement to the Board, she said the words attributed to her "run contrary to everything [she] stands for and significantly impugned [her] integrity," and that the adverse impacts of the video were, in her words, significant and quite traumatic.

Two people, including the councillor, reported the video under Meta's Bullying and Harassment policy. The Board's decision says Meta's systems did not prioritize the post for human review, and Meta's case page says it left the content up. Meta told the Board it did not remove the video because no Trusted Partner organization flagged it as false and likely to contribute directly to the risk of imminent harm, and because it was not election interference, per the Board's case-selection blog. Meta also concluded the video did not merit an AI label.

The Board found the video violated Meta's Hateful Conduct rules, because it alleges serious criminality and predatory sexual behavior against refugees as an entire group rather than against some of them, and should have been removed. It also found the video should have carried a "High Risk AI" label, and that Meta's rules on labeling deepfakes are inadequate.

The second case, 2026-043-FB-UA, concerns a Facebook post built on a manipulated TV interview. The Board says a young Muslim woman in Europe was interviewed as a campaign volunteer for menstrual health education, and that accounts across social media later manipulated the footage to ridicule her. The post in the case falsely presents her as her country's new health ambassador. For the user who appealed, both the initial report to Meta and the appeal were automatically closed. Meta told the Board that leaving the post up without an AI label was consistent with its policies, and it removed four disparaging comments only after the Board asked about them.

The Board overturned Meta there as well, requiring removal under the Bullying and Harassment policy. It found Meta's definition of "unwanted manipulated imagery" too narrow, because it covers changes to a person's appearance but not the falsifying of their words and actions.

Across the two cases the Board issued nine recommendations. They include lowering the threshold for "High Risk AI" labels, adding a click-through screen before users view labeled content, demoting or demonetizing that content, escalating penalties for accounts that repeatedly share it, publishing annual figures on how often labels are applied, and broadening the manipulated imagery definition to cover deepfakes of private individuals. The Guardian counted nine.

The recommendations are not binding. Under the Board's charter, it has authority to uphold or overturn Meta's content decisions, and Meta must respond to recommendations within 60 days. The Board advises Meta; it is not a court, and its rulings interpret Meta's own Community Standards, not the law.

Meta said it welcomed both decisions. "Meta will comply with the Board's decision within 7 days," the company said on its transparency pages. It did not commit to the policy recommendations. Oversight Board Co-Chair Pamela San Martin said the two cases showed "a broader, troubling pattern in which women who engage publicly on issues are disproportionately subjected to harassment and misinformation," per AFP.

The Board has pushed Meta on AI content before. It reiterated a recommendation from its AI-Generated Video in Israel-Iran Conflict decision, calling on Meta to publish a clear explanation of the penalties for failing to self-disclose digitally created or altered content.

For creators, both cases turn on the same gap: a synthetic video of a real person can spread without a label unless the uploader discloses AI use or a third party flags it. Meta's AI labels are a small banner reading "AI Info," Engadget notes, and the company has made few changes to them.

Meta must answer the nine recommendations within 60 days of the September 17 decisions, a window that runs to mid-November 2026.