Oversight Board urges Meta to lower AI label threshold and increase penalties

Meta's Oversight Board has overturned the company's decision to leave up an AI-generated video that contained hate speech targeting refugees. The board ordered the video's removal on September 17, and also instructed Meta to lower the threshold for applying 'high risk' labels to manipulated media, implement a click-through screen before users view such content, and penalize accounts that repeatedly share it.

The board found Meta's criteria for its 'High Risk AI' label to be excessively high, noting that the labels have been applied at a minimal scale. It also criticized Meta's enforcement guidance on hate speech as 'overly mechanical,' focusing on grammar rather than intent or potential harm.

The video, posted in November 2025, depicted a Scottish Labour councillor making inflammatory remarks about refugees. Meta had initially concluded that the video did not violate its community standards and did not warrant an AI label, despite user reports and appeals. The Oversight Board brought the case to Meta's attention.

Meta is now urged to expand the application of 'high risk' labels beyond crisis or election integrity measures, requiring an interstitial screen for viewers and imposing penalties like demonetization or removal from recommendations for repeat offenders. The board also called for globally applicable rules, increased transparency on label usage, and researcher access to content libraries.

The board rejected Meta's justification for allowing the video, stating that the absence of a crisis did not negate the content's potential to deceive the public. Meta did not penalize the user for failing to disclose AI use, and the company has previously resisted board recommendations to alter its interpretation of group-related statements as hate speech.