Meta to Label AI-Generated Content Across Facebook, Instagram

Meta will label AI-generated content on its platforms including Instagram and Facebook as part of changes to its content policies.

Starting in May, AI-generated user content will feature a “Made with AI” label to improve transparency and provide additional context to other users.

Detection of AI-altered content will come from the “detection of industry-shared signals of AI images,” Meta said.

Users can self-disclose if their content is AI-generated. Content deemed more likely to deceive users will receive a more prominent label. 

Meta previously removed manipulated content but will stop takedowns starting in July. The two-month window of May to July “gives people time to understand the self-disclosure process before we stop removing the smaller subset of manipulated media,” according to Monika Bickert, Meta’s vice president of content policy.

Meta will, however, still remove manipulated content if it violates the platform’s Community Standards on voter interference, bullying, harassment and incitement of violence.

Meta’s change in policy follows recommendations from its Oversight Board, which makes content moderation decisions for the company’s social media platforms.

Meta’s Oversight Board called the company’s current approach to AI content “too narrow” having only previously applied to video content.

Related:Meta Develops Invisible Watermarks to Track AI Image Origins

“The policy’s application to (i) only video content; (ii) only content generated or altered by AI; and (iii) content that makes people appear to say ‘words they did not say,’ is too narrow to meet any conceivable objective,” the Oversight Board wrote in February.

Meta’s new moderation approach now covers AI-generated video, audio and images.

Meta's Oversight Board’s comments on scope were made during a review of altered footage of President Joe Biden. The board also expressed concerns that taking down such content could impact freedom of expression.

“We agree with the Oversight Board’s recommendation that providing transparency and additional context is now the better way to address manipulated media and avoid the risk of unnecessarily restricting freedom of speech, so we’ll keep this content on our platforms so we can add labels and context,” Bickert said.

Meta also conducted research, asking 23,000 users about their thoughts on AI-generated content. Of those surveyed, 82% said they favored labels for AI-generated content.

Meta’s image generation feature, currently only available to U.S. users, already includes an “imagined with AI” watermark in the corner of the images it creates.

Related:Meta Advances Responsible AI With DINOv2 Release & New FACET Benchmark