Meta said on 6 February 2024, in a post by Nick Clegg, President of Global Affairs, that it would start labelling AI-generated images on Facebook, Instagram and Threads when industry-standard indicators are detected.
Meta already labels photorealistic images created with its own Meta AI generator as “Imagined with AI”, using visible markers plus invisible watermarks and metadata.
It is building detection for C2PA and IPTC “AI generated” signals so it can label images from tools including Google, OpenAI, Microsoft, Adobe, Midjourney and Shutterstock as those vendors add metadata.
For AI video and audio, where industry markers are less common, Meta said it would require users to disclose digitally created or altered photorealistic video or realistic-sounding audio, and may penalise failures to disclose.
Clegg framed the rollout as covering the following year of major elections, while Meta continues work on classifiers and harder-to-strip watermarks such as FAIR’s Stable Signature research.