Meta is testing facial recognition as a new layer in its fight against scam ads and account takeovers on Facebook and Instagram. The company says the technology is being used in limited ways: to identify suspicious celebrity-based ads, check possible impostor accounts and help locked-out users prove they own an account.
The move puts facial recognition back at the center of a familiar tradeoff. Meta is presenting the tests as a security tool for stopping fraud, while privacy questions remain especially sharp in places with stricter data protection rules.
How the scam ad test works
The main target is a scam format often called celeb-bait. In these ads, fraudsters use the image of a public figure, such as a content creator or celebrity, to draw people into clicking. The destination can be a scam website that asks users to share personal information or send money.
Meta already reviews ads with automated systems, including machine learning classifiers. The facial recognition test is meant to act as an additional check when existing systems flag an ad as suspicious and the ad appears to contain the image of a public figure at risk of being misused.
In that situation, Meta says it will try to compare the face in the ad with the public figure’s Facebook and Instagram profile pictures. If the company confirms both that there is a match and that the ad is a scam, it says the ad will be blocked.
This distinction matters because public figures also appear in legitimate advertising. Meta’s stated challenge is not simply finding a famous face in an ad. It is deciding when that image is being used to trick people into engaging with fraud.
What Meta says about facial data
Meta says the facial recognition feature for scam ads is not being used for any other purpose. According to the company, any facial data generated from an ad for the one-time comparison is deleted immediately, whether the system finds a match or not.
The company has described early tests with a small group of celebrities and public figures as promising. It did not specify who was included in that group. Meta also told TechCrunch it believes the same approach would be effective against deepfake scam ads, where generative AI has been used to create imagery of famous people.
In the coming weeks, Meta says it will begin showing in-app notifications to a larger group of public figures who have been hit by celeb-bait. Those public figures will be told they are being enrolled in the protection, and they can opt out in Accounts Center at any time.
Meta is also testing facial recognition for celebrity impostor accounts. In that case, the system would compare the profile picture on a suspicious account with a public figure’s Facebook and Instagram profile pictures. The goal is to identify accounts that scammers may use to impersonate public figures and create more opportunities for fraud.
Account recovery is the second test
Meta is also trialing facial recognition for account unlocking. This test is aimed at people who have been locked out of Facebook or Instagram after an account takeover, including cases where scammers trick someone into handing over passwords.
The process uses a video selfie. A locked-out user uploads the video, and Meta uses facial recognition to compare it with profile pictures on the account the person is trying to access.
Meta is positioning this as a faster and easier option than uploading an image of a government-issued ID, which is the usual route for restoring access now. The company says video selfie verification takes a minute to complete and argues that it will be harder for hackers to abuse than traditional document-based identity verification.
Meta also says the uploaded video selfie will be encrypted and stored securely. It says the video will not be visible on the user’s profile, to friends or to other people on Facebook or Instagram. As with the ad test, Meta says facial data generated from the comparison is deleted immediately whether there is a match or not.
The company compares the process to identity checks used to unlock a phone or access apps, including Apple’s Face ID on the iPhone. Still, the test asks users to trust Meta with a biometric recovery flow at the same time the company is under scrutiny for how it handles user data.
Why the timing is sensitive
Meta has faced years of criticism over scammers misusing famous people’s faces in ads, including scams around dubious crypto investments. The new tests are framed as a response to that long-running problem and to newer risks created by generative AI.
At the same time, TechCrunch notes the timing is notable because Meta is also trying to gather user data to train its commercial AI models as part of the wider industry push around generative AI tools. That makes the company’s narrow claims about the purpose of these facial recognition tests especially important.
The tests are being run globally, according to Meta, but not currently in the U.K. or the European Union. Those regions have comprehensive data protection regulations, and the source notes that, for biometric identity verification in the bloc, explicit consent is required from the individuals concerned.
Meta spokesman Andrew Devoy told TechCrunch that the company is engaging with the U.K. regulator, policymakers and other experts while testing moves forward. He also said Meta will keep seeking feedback and adjust the features as they evolve.
The core question
The security argument is straightforward: scammers misuse public faces, fake accounts and compromised logins, and Meta wants stronger tools to stop them. Facial recognition could help identify some of those abuses faster than existing systems alone.
The privacy question is just as direct. A tool built for a narrow fraud-prevention purpose can still raise concern when it depends on biometric comparison, especially from a company with broader ambitions in AI and digital identity.
For now, Meta’s position is that these tests are limited, purpose-specific and designed with immediate deletion of generated facial data. The next test will be whether public figures, locked-out users, regulators and privacy experts accept that boundary as the features expand.