Hardware Ready Ready to Bench?
Meta board deepfake removal. The Meta Oversight Board has ordered the removal from Facebook of two videos generated with artificial intelligence, targeting a Scottish Labour councilor and a young volunteer involved in a menstrual health campaign. Meta’s Oversight Board judged existing rules “constantly and profoundly inadequate” to address the rapid proliferation of synthetic content on its platforms.
In both cases, Meta had initially left the videos online, decisions later overturned by the board. The two proceedings highlight a worrying trend in using deepfakes to harass women and limit their participation in public debate, as emphasized by Pamela San Martin, co-chair of the board.
Meta board deepfake removal: why it matters
The first video, published on Facebook in November 2025, attributed a statement to the councilor that she had never made: “Refugees are welcome here, even if they rape our women, because whites do it too.” The audio presented imperfect synchronization with facial movements, suggesting manipulation. The footage was part of an album containing photographs of various women identified as participants in a demonstration against the far right and had surpassed five thousand views.
Two people, including the councilor herself, had reported the content. Meta’s automatic systems did not forward the video to a human reviewer, rejecting both appeals. The councilor reported having also filed a police report without results and described the use of her replicated voice as “rather traumatic.”
Meta had argued that the video violated no rules and did not require an AI-generated content label, a position maintained even after the board opened the case. The company explained that no organization in the Trusted Partner program had flagged the footage and that the content did not contain prohibited claims about voting or census.
The majority of the board concluded otherwise, establishing that the video violated rules against hate conduct because it attributed “serious criminality and predatory sexual behavior” to refugees as a whole group, rather than to individual persons. The board recommended applying the “High Risk AI” label. A minority of the board held that the primary target was the positions attributed to the politician, not refugees themselves; regarding bullying and harassment, no violation was found given the councilor’s status as a public figure.
The second proceeding concerned a young Muslim woman in Europe who had participated in a television interview as part of a menstrual health education campaign aimed at women and girls belonging to ethnic minorities. After the broadcast, accounts on various platforms digitally manipulated images from the interview to mock her, generating videos that accumulated tens of millions of views, also on Meta’s platforms.
What changes and what are the effects
The board focused on one of these videos, which depicts her performing absurd exercises and consuming junk food, judging it a violation of bullying and harassment rules. The report of the video and subsequent appeal were automatically closed by Meta’s systems without applying any label regarding synthetic origin. The company removed harassing comments under the post only after the board’s intervention.
The board formulated nine recommendations: six related to the first case and three to the second. Among these, lowering the threshold for activating the “High Risk AI” label, currently linked to crises and elections, and pairing or replacing it with a warning screen requiring a click before viewing. The board also requests that algorithms reduce the distribution of content classified as high risk, that such content be demonetized, that repeat offender accounts face stricter sanctions, and that Meta annually publish data regarding the application of labels.
Other recommendations concern the review queue, prioritizing potentially synthetic content once reported, and expanding the definition of “unwanted manipulated images” to include deepfakes showing a private citizen saying or doing things they did not say or do. Decisions on individual content are binding for Meta, which has communicated the removal of the two videos. Recommendations regarding policy are not: the company has sixty days to indicate its intentions and in the past has accepted only few of the board’s proposals on labeling synthetic content. The tool currently in use is the “AI Info” banner.
Source and further reading on Meta board deepfake removal: original article.
* Content created with the assistance of artificial intelligence systems.