Oversight board wants Meta to automatically label more AI-generated content

Meta’s enforcement of its manipulated media policy is inconsistent, incoherent, and unjustifiable. Therefore, it must invest in technology that can identify and label AI-manipulated audio and video at scale so users are properly informed.

That’s the conclusion of Meta’s oversight board, an advisory body that independently advises on content moderation decisions on Facebook, Instagram, and Threads.

Meta began labeling AI-generated content on its social media platforms a year ago. Users could flag a post or video when it was generated using AI technology. However, the American tech company also promised to deploy a detection system to proactively flag AI-generated content. This was all documented in Meta’s manipulated media policy.

Attempts to automatically apply a label to all instances of manipulated media haven’t been successful, Meta’s oversight board says.

The Board mentions a case in which Meta decided not to label a manipulated audio clip of two Iraqi Kurdish politicians discussing rigging Kurdistan’s parliamentary elections in October 2024. The media outlet’s Facebook page had about four million followers. The post had about 200,000 views.

Two users reported the content as misinformation, but Meta decided to keep the audio clip online. One of the users appealed to Meta, but the company upheld its decision based on a classifier score. The user then filed a complaint with the oversight board, which in turn overturned Meta’s decision not to label the audio clip.

“When it comes to identifying AI-created or manipulated content on its platforms, Meta told the board that it is only able to automatically identify and label static images, not video or audio content. Given the company’s expertise and resources and the wide usage of Meta’s platforms, it must prioritize investing in technology to identify and label manipulated video and audio at scale,” the oversight board recommends.

In addition, the advisory board is concerned that Meta’s manipulated media labels are not available in every local language on Meta’s platforms, including Sorani Kurdish. The board stipulates that this should be part of the company’s electoral integrity efforts.

Therefore, Meta should ensure that the informative labels for manipulated media on Facebook, Instagram, and Threads are displayed in the same language that the user has elected for its platform.

Lastly, the oversight board is also concerned by Meta’s reliance on third parties for technical assessments of likely manipulated content.

“Meta should reconsider having this expertise available internally,” the board concludes.