Board orders removal of deepfakes
Meta has been instructed by its oversight board to remove from Facebook two sets of AI-manipulated videos — one featuring a British politician, the other a young Muslim woman — and to do far more about synthetic fake imagery. According to the board, a fabricated clip of a Scottish Labour councillor voicing inflammatory remarks about refugees ought to have been taken down. The board likewise concluded that an AI-made video of a Muslim campaign volunteer, which falsely showed her giving health tips while doing ridiculous exercises or eating junk food, should have been deleted.[S1]
The board characterised Meta's protections as "consistently and fundamentally inadequate" in the face of fast-growing AI deepfakes. Oversight board co-chair Pamela San Martin noted that such synthetic videos are increasingly weaponised to harass women and drive them out of public debate. According to her, these two cases reveal a wider and worrying trend: women who speak publicly on issues face harassment and false information at disproportionate rates, and Meta along with other platforms must adopt stronger rules to curb the spread of deepfakes.[S1]
Scotland councillor video
The Scottish video falsely attributes to the councillor a claim that refugees are welcome even if they rape women, since white people do so as well. The oversight board, whose rulings bind the company and which operates at quasi-arm's length, judged the clip to be AI-generated, pointing to audio that did not line up properly with the unnamed councillor's facial movements. Yet when the board brought the matter directly to Meta, the company concluded the video breached no content rules and warranted no AI label.[S1]
Writing in a blogpost on Thursday, the board argued the post warranted removal because it broke Meta's hateful conduct rules by portraying refugees as a whole, rather than as individuals, as criminals and sexual predators. The board added that the video should have carried a "high risk AI" tag, and pressed for stricter deepfake measures. A board majority concluded Meta's deepfake policies need strengthening: broadening when "high risk" labels apply, doing more to curb the reach of deceptive AI material, raising penalties for accounts that repeatedly post it, and publishing more data on when AI labels get applied.[S1]
Nine policy changes recommended
The board put forward nine changes to Meta's policies, among them a requirement that algorithms downrank material carrying the "high risk" label so it surfaces less often in users' feeds. It further urged Meta to make AI-generated content harder to reach, suggesting measures such as an interstitial warning screen that users must click through before they can view the post.[S1]
In a second decision, the board told Meta to delete from Facebook an AI-made video that ridiculed a young Muslim woman in Europe involved in a campaign for menstrual health education and against stigma surrounding health conditions affecting women and girls from ethnic minority communities. Manipulated clips and images mocking her drew tens of millions of views online, Meta's platforms among them. The board flagged one video — in which she exercises absurdly and eats junk food — as violating Meta's bullying and harassment rules. It called for removal and issued three policy recommendations, among them widening the definition of "unwanted manipulated imagery" so it covers deepfakes of private individuals shown saying or doing things they never said or did.[S1]
Background and response
Meta established the board in 2020 to act as an arbiter for content hosted on its platforms. Meta has been contacted for a response.[S1]
