Meta’s Oversight Board is investigating an apparently AI-generated video of a British politician that misrepresents their views on issues such as immigration. The board said the video in question was part of an album posted on Facebook in November last year, and features a seemingly deepfaked Labour Party councillor who represents an area in Scotland making sarcastic and offensive comments about refugees and sexual assault.
As well as the deepfake video, the post also includes a video of pro-Palestine protesters that the Oversight Board also suspects to be AI-generated, plus a likely genuine still image of several women, including the politician from the first video, holding anti-far-right signs. The women are also named in the post, and the written caption is said to make baseless accusations of tax evasion against the Labour politician.
The board says engagement with the post was relatively minimal, but it contained no AI labels or disclosures. It added that two users had alleged that the content violates Meta’s Bullying and Harassment policy, but when the company’s systems failed to escalate their complaints for human review and the content remained on Facebook, one of them appealed to the board.
The case outline goes on to explain that Councillor in question has been vocal about what they believe to be racist misinformation weaponized to fuel anti-migrant sentiment amid protests against the housing of asylum seekers in UK hotels. They have previously reported threats and intimidation tactics used against them for speaking out on the issue, which have included AI-generated online defamation.
The Oversight Board claims Meta said the post was not flagged for removal or perceived to be in violation of its rules because the politician is a public figure of adult age, and therefore not automatically protected from “unwanted manipulated imagery.” Private individuals can self-report harmful content and request to have it taken down.
Meta also objected to claims that the content made generalized statements about refugees that were in violation of its Hateful Conduct policy, and said it didn’t meet the criteria for misinformation as it hadn’t been reported by any member of its “Trusted Partner” program. It was also not deemed to be election interference as the councillor portrayed did not stand in Scotland’s recent local elections, and Meta referred to the low engagement and “satirical” nature of the video as reasons why it was not required to have an AI label applied to it.
The Oversight Board is now inviting public comments relevant to the case, which may inform policy recommendations it might make to Meta. These are not binding, but Meta is required to respond to them within 60 days. As a reminder, the Oversight Board is an independent external body despite its ultimate financial reliance on Meta.
Meta’s apparent indifference to the deepfake of the Scottish politician is even more surprising when you take into account the hot water it found itself in last year concerning an AI-generated video of the (successful) Irish presidential candidate Catherine Connolly. In the video, the now President of Ireland appeared to announce that she was withdrawing from the election campaign, and it was shared nearly 30,000 times on Facebook before being removed.
Connolly called the video “a disgraceful attempt to mislead voters and undermine [Ireland’s] democracy,” and Meta swiftly killed the account it was hosted on after being contacted by the Irish Independent. In this case, the video was ruled to be in breach of Meta’s community standards, and it was far from the first time that it has struggled to deal with the emergence of politically-motivated deepfakes.
