Meta was told to label any media that featured edits of politicians generated by artificial intelligence.

The Oversight Board, an organization created by Meta to oversee what content the company removes from Facebook and Instagram, ruled on Monday that Meta had to leave AI-generated media designed to deceive the public on its platforms but that it had to label them as such.

This ruling focused on a video posted to Facebook in May 2023 in which the user edited the footage to make it look like President Joe Biden was inappropriately touching his granddaughter. The unaltered footage featured Biden giving his granddaughter an “I voted” sticker.

Meta initially decided that the footage did not breach its rules since it was not a “deepfake,” which it considers to be a video in which the speaker’s audio was modified. The board upheld Meta’s decision on the grounds that the video did not show Biden “saying words he did not say” or “doing something he did not do” but implored the company to update its “incoherent” and “confusing” policy on deepfakes.