The Oversight Board calls Meta’s uneven AI moderation ‘incoherent and unjustifiable’

As Meta’s platforms fill up with more AI-generated content, the company still has a lot of work to do when it comes to enforcing its policies around manipulated media. The Oversight Board is once again criticizing the social media company over its handling of such posts, writing in its latest decision that its inability to enforce its rules consistently is “incoherent and unjustifiable.”

If that sounds familiar, it’s because this is since last year the Oversight Board has used the word “incoherent” to describe Meta’s approach to manipulated media. The board had previously urged Meta to update its rules after a misleadingly edited video of Joe Biden went viral on Facebook. In response, Meta said it its use of labels to identify AI-generated content and that it would apply more prominent labels in “high risk” situations. These labels, like the one below, note when a post was created or edited using AI.

This approach is still falling short though, the board said. “The Board is concerned that, despite the increasing prevalence of manipulated content across formats, Meta’s enforcement of its manipulated media policy is inconsistent,” it said in its latest decision. “Meta’s failure to automatically apply a label to all instances of the same manipulated media is incoherent and unjustifiable.”

The statement came in a decision related to a post to feature audio of two politicians in Iraqi Kurdistan. The supposed “recorded conversation” included a discussion about rigging an upcoming election and other “sinister plans” for the region. The post was reported to Meta for misinformation, but the company closed the case “without human review,” the board said. Meta later labeled some instances of the audio clip but not the one originally reported.

The case, according to the board, is not an outlier. Meta apparently told the board that it can’t automatically identify and apply labels to audio and video posts, only to “static images.” This means multiple instances of the same audio or video clip may not get the same treatment, which the board notes could cause further confusion. The Oversight Board also criticized Meta for often relying on third-parties to identify AI-manipulated video and audio, as it did in this case.

“Given that Meta is one of the leading technology and AI companies in the world, with its resources and the wide usage of Meta’s platforms, the Board reiterates that Meta should prioritize investing in technology to identify and label manipulated video and audio at scale,” the board wrote. “It is not clear to the Board why a company of this technical expertise and resources outsources identifying likely manipulated media in high-risk situations to media outlets or Trusted Partners.”

In its recommendations to Meta, the board said the company should adopt a “clear process” for consistently labeling “identical or similar content” in situations when it adds a “high risk” label to a post. The board also recommended that these labels should appear in a language that matches the rest of their settings on Facebook, Instagram and Threads.

Meta didn’t respond to a request for comment. The company has 60 days to respond to the board’s recommendations.