Meta’s Oversight Board overturned the company’s decision to leave a deepfake of a Scottish councillor on Facebook, finding that it should have been removed under Meta’s Hateful Conduct rules. The Board also called for stronger friction, penalties and transparency around high-risk AI content.
Meta AI Watch analysis
What happened
The video, posted in November 2025, falsely depicted the councillor making a hateful statement about refugees. Genocide Watch’s account of the Board’s 17 September decision says the post remained up after two users reported it and appealed to Meta. The Board required its removal.
The Board’s majority also said the video should have received a “High Risk AI” label, while a minority disagreed about whether it warranted removal or a different label. The Board recommended that Meta lower the threshold for high-risk labels, put an interstitial in front of labelled content, restrict its reach or monetisation, and increase penalties for repeat sharers. It also called for annual disclosure of label use, including on content from or about politicians. Genocide Watch’s account of the decision summarises the ruling and recommendations.
Why it matters
This is not just a debate about whether a synthetic video carries a label. The Board’s recommendations would add consequences around viewing, distribution and repeat sharing, while making it easier to see how often Meta applies its strongest AI labels.
The case also exposes a difficult boundary: the Board majority treated the video as hate speech targeting refugees, while a minority saw it as political speech about the councillor’s alleged views. That disagreement matters when platforms decide whether a realistic fabrication should be removed, labelled or left up.
Our read
A label that viewers barely notice is a rather modest answer to a video that can put hateful words in a real person’s mouth. The Board has laid out more practical options, but they are recommendations, not evidence that Meta has adopted them. The meaningful test is whether the company changes its rules and shows how consistently it applies them.
What to watch
- Whether Meta changes its high-risk AI labelling threshold or adds an interstitial.
- Whether it limits recommendations or monetisation for labelled content and repeat sharers.
- Whether Meta publishes the label-use data the Board requested.
- How future cases distinguish hate speech from political speech involving synthetic media.
Discussion spark: Should a realistic AI deepfake that is labelled but not removed be treated as adequately moderated, or should platforms restrict its reach and penalise repeat sharing?
Sources and evidence
- AI and Hate Speech: Meta Must Do More Against Harmful Deepfakes Containing Hate Speech – Genocide Watch (5 October 2026, 23:05 UTC)
not affiliated with, endorsed by, or operated by Meta