(Credit: Koshiro K/Shutterstock)
Meta's Oversight Board is investigating some of the tech firm's policies and past actions surrounding AI-generated nude images and deepfakes of women, citing two separate incidents where AI-generated naked images of real-life women were posted to Instagram and Facebook.
In a Facebook group for AI-generated content, a nude post simulating a female American public figure being groped by a man was taken down. The board is not revealing the identity of the woman in an effort to protect her privacy and prevent harassment. This offending image was added to a media-matching database so that the image would automatically be detected if reposted elsewhere on a Meta-owned platform. The user who posted the explicit AI creation tried to appeal the takedown, but failed, and then tried to appeal to the board itself.
On Instagram, an account that only posts AI-generated images of women of Indian descent posted a nude deepfake of a female public figure from India, whose name was also concealed for privacy reasons. After a user reported the image, the report was closed because no one at Meta reviewed it within two days. After the user appealed the inaction, the appeal was subsequently shut down as well.
The board said that only after it flagged this incident to Meta did the company remove this deepfake, citing its "Bullying and Harassment" policy.
These two separate incidents suggest an inconsistent enforcement policy regarding AI-generated deepfakes of real women. It's unclear why Meta would automatically close reports if unreviewed, and it's also unclear why the post of the American woman was quickly taken down while the image of the woman from India was not. It's also unclear why one image was added to the detection database, while the other seemingly was not. PCMag has reached out to Meta for comment.
"The Board selected these cases to assess whether Meta’s policies and its enforcement practices are effective at addressing explicit AI-generated imagery," the group wrote in its post announcing the investigations.
For better or worse, AI-generated and manipulated media are allowed on Facebook and Instagram, though nudity or sexually explicit content is generally not allowed. Tons of AI-generated, spammy images have recently gone viral on Facebook, like "Shrimp Jesus."
In February, Meta shared that it will begin attempting to label AI-generated content. Meta later said it will expand the scope of content it will flag next month and plans to attach "Made With AI" labels to more posts that don't already contain "invisible" data notating AI origins.
Earlier this year, Meta's Oversight Board decided that a manipulated video of President Joe Biden that made him look like he was engaging in illicit behavior could stay on Facebook. The board did say at the time that Meta's manipulated media policies were lacking, though.
Back in January, Taylor Swift deepfakes swirling online sparked conversations around how famous figures can get priority treatment compared to non-famous deepfake victims. In the case of Swift, Twitter/X temporarily put a site-wide search ban on the star's name in an effort to quell the flood of fake images.
Some countries are taking action at the government level to stop the spread of deepfakes, which primarily target women. This week, UK's Ministry of Justice has shared that making sexually explicit deepfakes will soon be illegal in the country, The Guardian reports.
In the case of Meta, its Oversight Board plans to debate the aforementioned cases of the two unnamed American and Indian women "over the next few weeks" before publishing final determinations.