Meta Resumes Facial Recognition Testing
By Byron Kaye and Katie Paul
SYDNEY/NEW YORK (Reuters) – Three years after shutting down its facial recognition software due to privacy concerns, Meta is back testing the service as part of a crackdown on “celeb bait” scams.
Meta plans to enroll about 50,000 public figures in a trial that will automatically compare their Facebook profile photos with images used in suspected scam advertisements. If they match, Meta will block the ads.
Celebrities will be notified and can opt out if they prefer not to participate.
The global rollout of the trial is set for December, although it will exclude some regions like Britain, the EU, South Korea, and certain U.S. states that lack regulatory clearance.
Monika Bickert, Meta’s VP of content policy, emphasized the goal of providing protection to public figures whose likenesses have been exploited in scams. “The idea here is to roll out as much protection as we can for them,” she stated.
This approach aims to balance the use of invasive technology to address rising scam numbers while minimizing user data complaints—a persistent issue for social media platforms. When Meta shut down its facial recognition system in 2021, it cited “growing societal concerns” and erased data on over one billion users. Earlier this year, it paid Texas $1.4 billion over a lawsuit regarding illegal biometric data collection.
Additionally, Meta faces lawsuits claiming it hasn't done enough to prevent celeb bait scams, which often utilize AI-generated images of celebrities to mislead users into investing in fake schemes.
Under the new trial, any face data created during comparisons will be deleted immediately, regardless of whether a scam is detected.
Bickert stated that the tool underwent a thorough privacy and risk review process and involved discussions with regulators and privacy experts before testing began.
Meta is also exploring using facial recognition data to help non-celebrity users regain access to compromised accounts on Facebook and Instagram.
(This story has been refiled to fix a typo in paragraph 2)
Comments (0)