Meta rolls out new labeling system for AI-generated content

Meta announced Friday that it is making changes to the way it handles content generated by artificial intelligence. The company will begin labeling a broader range of photos, audio and video as being “made with AI” on Facebook, Instagram and Threads.

“We’re taking steps we think are appropriate for platforms like ours. We want to help people know when photorealistic images have been created or edited using AI,” Meta Vice President of Content Policy Monica Bickert said in a statement on the company’s website posted Friday.

The new label will be used when Meta detects indications of an AI-generated image or when people voluntarily disclose they have uploaded content generated with AI. The decision was made following a recommendation from the company’s oversight board that it needed to provide transparency to “avoid the risk of unnecessarily restricting freedom of speech.”

Meta’s existing policy was written in 2020, before AI-generated content was widely used. It only covers videos created or altered by AI to make a person appear to say something they did not. Recognizing the policy was too narrow when AI was evolving to generate audio and photos, Meta said its new policy addresses AI-generated content that shows a person doing something they did not do.

Already, Meta web sites labels photorealistic images created with its Meta AI feature as “imagined with AI.”

Bickert said Meta began looking at its policies last spring “to see if we needed a new approach to keep pace with rapid advances in generative AI technologies and usage.”

After consulting with over 120 stakeholders in 34 countries, she said there was widespread support for labeling content generated with AI and for more prominent labeling in high-risk scenarios. A majority of consultants said content removal was warranted in a limited number of “highest-risk scenarios where content can be tied to harm, since generative AI is becoming a mainstream tool for creative expression.”

Bickert said Meta also conducted public opinion research with 23,000 people in 13 countries to ask them how media companies, including Meta, should handle AI-generated content. It found that 82% favored warning labels for AI-generated content depicting people saying things they did not in fact say.

Bickert said that Meta will continue to review is approach to AI as technology progresses, doing so in collaboration with peers in the industry and conversations with government and members of the public.