Meta, owner of Facebook and Instagram, announced major changes to its policies on digitally created and altered media on Friday, before elections poised to test its ability to police deceptive content generated by artificial intelligence technologies.
The social media giant will start applying “Made with AI” labels in May to AI-generated videos, images and audio posted on Facebook and Instagram, expanding a policy that previously addressed only a narrow slice of doctored videos, the vice-president of content policy, Monika Bickert, said in a blogpost.
Bickert said Meta would also apply separate and more prominent labels to digitally altered media that poses a “particularly high risk of materially deceiving the public on a matter of importance”, regardless of whether the content was created using AI or other tools. Meta will begin applying the more prominent “high-risk” labels immediately, a spokesperson said.
The approach will shift the company’s treatment of manipulated content, moving from a focus on removing a limited set of posts toward keeping the content up while providing viewers with information about how it was made.
Meta previously announced a scheme to detect images made using other companies’ generative AI tools by using invisible markers built into the files, but did not give a start date at the time.
A company spokesperson said the labeling approach would apply to content posted on Facebook, Instagram and Threads. Its other services, including WhatsApp and Quest virtual-reality headsets, are covered by different rules.
The changes come months before a US presidential election in November that tech researchers warn may be transformed by generative AI technologies. Political campaigns have already begun deploying AI tools in places like Indonesia, pushing the boundaries of guidelines issued by providers like Meta and generative AI market leader OpenAI.
In February, Meta’s oversight board called the company’s existing rules on manipulated media “incoherent” after reviewing a video of Joe Biden posted on Facebook last year that altered real footage to wrongfully suggest the US president had behaved inappropriately.
The footage was permitted to stay up, as Meta’s existing “manipulated media” policy bars misleadingly altered videos only if they were produced by artificial intelligence or if they make people appear to say words they never actually said.
The board said the policy should also apply to non-AI content, which is “not necessarily any less misleading” than content generated by AI, as well as to audio-only content and videos depicting people doing things they never actually said or did.