Meta silently added facial recognition code for its smart glasses to millions of phones

According to the online analysis of the company's software,Meta. Human face recognition of its smart glasses has been quietly embedded in an application that has been downloaded to millions of mobile phones.

After several updates this year, the code carefully added in Meta's artificial intelligence application shows that this feature (internally known as “NameTag”) can identify the person caught in the eyeglass camera and, when activated, alerts the person to wear when the person is identified.

The name Tag found in a real-time Meta AI application indicates that Meta has started sending face recognition codes to the user's mobile phone, while publicly describing the company as still “thinking”. In April this year, Meta indicated that if the face recognition technique was to be used, it would not be introduced without first taking a “very thoughtful approach”. However, Connect found that, as early as January, the core components of the system had been integrated into software distributed to millions of people.

Although not yet enabled, the name Tag is in the Meta AI application, which has been downloaded more than 50 million times and is necessary to use its smart glasses (including the Ray-Ban and Oakley models). If activated, it converts the captured faces of Meta glasses into unique biometric signatures, commonly known as facial marks, and checks each signature against the face marks stored on the user ' s mobile phone, which is currently configured to receive updates from Meta. The identified face will trigger the notice, while the remaining face will be cut, indexed and stored in a folder marked " Pending " .

Zenium Need a tip?

|--|

You're a current or former employee of Meta. Want to talk about company technology? We would like to receive your letter. Journalists are safely contacted through Signal (dmehro.89 or dell.3030) using non-working telephones or computers. Zenium

Name Tag will revitalize Meta, a technology that was abandoned in 2021, when the company announced that it would remove more than 1 billion facial fingerprints belonging to Facebook users, after the company ' s photo tag system had been controversial. Meta eventually paid $650 million to settle a class action brought by Illinois users and agreed in 2024 to a separate settlement of $1.4 billion with Texas for its alleged illegal collection of user biometric data.

At a time of growing opposition to face recognition at the consumer level, the company has made new efforts, which, according to privacy advocates, will make dangerous technologies easily available to anyone from stalkers to immigration agents. The Meta internal document released in the New York Times in February showed that the company planned to introduce the function in a “dynamic political environment”, and Meta believed that the biggest critics would be focused.

According to the " Wiring " analysis (reproduced independently by external experts), three artificial intelligence models that support the name Tag have been deployed from Meta ' s server and are now located on the client ' s mobile phone. A model detects human faces, a model tailors them, and a third model encodes them into biometric data.

There are only traces of the user interface, suggesting how the functionality works eventually. The May version of the application renamed the user function " Connection " and invited them " to remember the person you met " . It is not clear which faces will be included in the system's identification database, how these personal data will be created or how many people can eventually be identified through it.

Connected magazine shared its findings with two external security researchers, who examined the application separately and reproduced key aspects of the analysis: Cooper Quintin, a security researcher and a senior public interest technical expert at the Non-Profit Electronic Front Foundation Threat Laboratory, and an independent safety and privacy researcher under the alias Butchodi, spent more than a decade working backwards on consumer software and surveillance techniques.

“The function has not yet been demonstrated to consumers, but appears to be ready,” said Quentin. “Despite the billions of reasons for not doing so, Meta seems to have created the capacity to turn its clients into a distributed monitoring machine.”

Buchodi performed additional tests on the identification pipeline. (Please read here their technical analysis.) In order to see whether the matching system is working, Boujodi added a face fingerprint from the late French philosopher Michel Foco to the chart of the application. When Fukuko's image triggers the name tag, the application produces a notice: “This person has been identified”.

“The main components of the face recognition feature are already contained in Meta’s companion application”, said Buchodi. "There's not much to block this function and work."

In April of this year, more than 70 advocacy groups, including the American Civil Liberties Union, the Electronic Privacy Information Centre and Futures International, asked Meta to abandon its name Tag, warning that it would allow stalking and abusers to identify strangers in public. In a statement to Connect at the time, the spokesperson for Meta said: “Our competitors provide this type of face recognition product, and we do not”. “If we were to publish such a function, we would have taken a very considered approach before launching anything.”

Privacy advocates believe that Meta could normalize functions previously withdrawn because of privacy problems by embedding face recognition in a dressable platform in the public market.

“You can set norms and standards by integrating technology into ecosystems,” said Joseph Jerome, former Meta Reality Labs policy officer, responsible for the company's AR and VR product privacy reviews, on the role of Meta in the wearable technology industry. "I don't know how Meta responsibly deploys this technology."

“Whatever the sensational reports are, the facts are simple: as we said before exploring these types of functions, what you see is evidence of such exploration,” says Meta spokesman Ryan Daniels. “Nothing has been shipped to consumers and no final decision has been made on what to do here, if any. If we really decide to introduce something, we will take a considered approach and proceed in a fully transparent manner. One of the clear decisions we can make — we will not create a central human face database.”

The Code Review of Connect indicates that the current design of the NameTag system is to extract facial fingerprints from Meta ' s server and store them on user devices.

The early Meta system, published on Facebook in 2010, can analyze photos and recommend labels for people who appear in user images. It rapidly spread to over 1 billion users and became one of the largest consumer facial recognition systems ever deployed.

This technology attracted close attention almost immediately. European regulators and privacy advocates in the United States questioned their legitimacy as early as 2011 and were concerned about the meaningful consent of users to the creation of biometric data. In 2019, Meta paid $5 billion to the Federal Trade Commission and the Ministry of Justice to address broader privacy cases, including face recognition.

In November 2021, Meta announced that it would close the system and remove its built human face template on the grounds that there was growing concern about the role of face recognition in society. However, Jerome, who joined Reality Labs in mid-2021, stated that the decision had never been understood internally as a permanent withdrawal. "There is always tension: when can we cancel face recognition?"

According to the internal documents consulted in The Times, Meta plans to introduce for the first time, in 2025, the face recognition feature of its smart glasses to the participants of the Blind Conference, which will then be open to the public. But never. But this technology has met the real need: the available assistive devices already allow blind users to identify their personal registered faces, and a study of blind users by Cornell University of Technology and Facebook researchers in 2018 found that each participant described the identification of people as an important daily task.

Meta did not answer the question as to which users could be identified by name Tag; whether it was intended to transfer back to its server the photographs, facial features or other data generated by the system; or whether the company planned to let the user choose to join rather than quit. Essilor Luxottica, who co-produced Ray-Ban and Oakley smart glasses with Meta, did not respond to the request.

Woodrow Hartzog, a professor of privacy law at Boston University, said that even opting for protection (if Meta eventually protects) would be weak. He said that consent could normally be linked to work, welfare or access to services. The issue of considering privacy as an individual choice is beneficial to the enterprise and does not impose any meaningful restrictions on collection, while allowing the company to claim control over the user.

“We know that the more these systems are deployed, the more people think they are normal,” Hazorg said. “The more we think they are ordinary, routine, the more people begin to judge the desirability or usefulness of facial scans on their own moral trail. That's the human mind."