Facebook said it will remove video that has been edited or synthesized “in ways that aren’t apparent to an average person and would likely mislead someone into thinking that a subject of the video said words that they did not actually say.” Monika Bickert, the company’s VP of global policy management, laid out the rule in a policy update in January 2020.
The ban also covers media made with artificial intelligence that merges, replaces or superimposes content onto a video so it looks authentic. TechCrunch said that will likely catch the crudest fakes, such as revenge deepfake porn that puts a real person’s face on an adult performer’s body, though nudity was already banned on the platform.
What stays up
The carve-outs are wide. Parody and satire are exempt. So are edits made for quality, and cuts or splices that only shorten a video or reorder words.
That leaves some well-known cases untouched. TechCrunch pointed to the doctored clip of House Speaker Nancy Pelosi that went viral on Facebook the year before, with audio slowed down to make her sound drunk or ill. The Washington Post suggested it would not qualify for removal under the new policy. Neither would an edit from the recent UK general election, in which one party’s campaign staff cut a video of a rival politician being asked about Brexit so he looked lost for words when he was not. The viral deepfake of Mark Zuckerberg speaking like a megalomaniac also looks safe, since its two artist creators called it satire.
TechCrunch summed the policy up as “no to malicious deepfakes but spindoctors please carry on.”
Fact-checkers and labels
Manipulated content that falls short of the removal standard can still go to Facebook’s independent third-party fact-checkers, who may rate it false or partly false. Bickert wrote that such a rating cuts its distribution in News Feed “significantly,” bars it as an ad, and puts warnings in front of anyone who sees, tries to share or has shared it.
She defended leaving the material up. “If we simply removed all manipulated videos flagged by fact-checkers as false, the videos would still be available elsewhere on the internet or social media ecosystem,” she wrote. “By leaving them up and labelling them as false, we’re providing people with important information and context.”
The business angle
The month before, Facebook announced it had found a network of more than 900 fake accounts spreading pro-Trump messaging, some using profile photos generated by AI. TechCrunch noted that AI-made faces could make that fake-account problem worse, which would be bad for business, and that this gives the company its own reason to act. It also noted that California had banned political deepfakes around elections that fall.
Bickert pointed to other work: a Deep Fake Detection Challenge announced the previous fall to produce research and open source detection tools, and free online training courses for journalists on spotting manipulated visuals, run with Reuters. “As these partnerships and our own insights evolve, so too will our policies toward manipulated media,” she wrote.
