Meta Glasses Backlash: Hidden Facial Recognition Code Found and Removed

Researchers found facial recognition code quietly embedded in the Meta AI companion app for the company's Ray-Ban smart glasses. EFF's Threat Lab verified it through static analysis. WIRED reported it. Less than 48 hours later, Meta pushed an update and the code was gone. The company has since refused to answer whether the feature could return, or what happened to any data collected during internal testing.

That last part is the story.

The feature, known internally as "Name Tag," was unactivated in the public app build. Users could not trigger it. But the infrastructure was complete: machine learning models, biometric databases, and logic ready to fire "Person recognized" alerts, per EFF. A June 5 update scrubbed all of it, EFF confirmed. Meta has not commented on what, if anything, the system collected while it was being tested internally.

How "Name Tag" worked and why dormant code still alarmed critics

The Meta AI app contained code to detect faces, convert them into unique biometric signatures, store those signatures in databases, and surface real-time information about identified individuals through Meta AI, per EFF and the Journal of High Technology Law. The tool would not have allowed unrestricted use of facial recognition, the Journal of High Technology Law noted earlier this year.

That distinction matters technically. What it doesn't resolve is the bystander problem. The proposed system would have operated continuously in public spaces, generating biometric identifiers for people who never interacted with Meta's products, never agreed to any terms of service, and had no practical way to opt out, as the Journal documented. A wearer walking through a grocery store, a courthouse, or a protest would still scan everyone in their line of sight, restricted feature set or not.

The stakes aren't abstract. A spokesperson for Refuge and Women's Aid, a domestic abuse charity, told the Journal that "[s]talking is an extremely common tactic used by perpetrators, and an instant identification feature could place survivors in harm's way by enabling abusers to locate and track them," as cited by the Journal of High Technology Law.

The "unactivated" framing is technically accurate and incomplete at the same time. The system was built. The intent was written down. And a launch strategy was already in circulation. An internal Meta memo, obtained by The New York Times and reported on by EFF and the Journal of High Technology Law, stated the company may launch "during a dynamic political environment where many civil society groups that we would expect to attack us would have their resources focused on other concerns." Privacy advocates read that line as a timing strategy calculated against distraction, not a product-safety guardrail.

Meta smart glasses backlash was building before the code was found

EPIC and EFF had already urged state attorneys general and the FTC to investigate Meta's reported plans before WIRED confirmed that functional code was present in the app, per the Journal of High Technology Law. Their letters warned the feature could "significantly expand surveillance risks and threaten civil liberties." When the code was verified, those warnings had a concrete object.

When WIRED published, Meta's executives went on the defensive, according to EFF. Then came the June 5 update. Then silence on every substantive follow-up question.

EFF is an advocacy organization, and its framing should be read as such. On the core sequence of events code found, code verified, code removed the documented facts track closely with its account.

A billion-dollar pattern

The speed of the rollback makes more sense with some financial history in view.

Meta shut down Facebook's photo-scanning face recognition tool in November 2021 and deleted more than one billion face templates, per EFF. That reversal came after years of regulatory pressure and expensive litigation: a $5 billion FTC settlement in 2019 over deceptive privacy practices; a $650 million settlement with Illinois consumers under the state's Biometric Information Privacy Act, which requires affirmative written consent before biometric data can be collected and gives individuals the right to sue directly; and a $1.4 billion settlement with Texas in 2024 over the same defunct system, all documented by EFF.

Smart glasses sharpen the legal exposure considerably. Unlike a social-media photo upload, the proposed Name Tag system would have scanned bystanders who never touched Meta's products. That profile fits poorly with Illinois-style biometric privacy statutes, and dozens of states now require affirmative consent before companies can collect biometric data, per EFF. EFF's characterization that Meta "stopped only after it faced the legal and financial consequences" in prior rounds is advocacy language. The settlement figures are not.

What the rollback doesn't settle

Meta has refused to answer WIRED's questions about whether it plans to bring Name Tag back, or what happened to any data that may have been collected during internal testing, per EFF. Those are the two questions that matter most, and neither has an answer.

The underlying architecture also remains intact. No AI features on the Ray-Ban Meta glasses run locally; all of them route through Meta's servers, according to EFF's consumer advisory from earlier this year. Removing the facial recognition code from the app does not change that infrastructure.

What this episode surfaces is a structural problem that a quiet app update cannot fix. Ambient, wearable surveillance directed at people who never opted into any Meta product is a different category of risk than social-media facial recognition, and the legal frameworks available to address it were not built for this scenario, as the Journal of High Technology Law concluded earlier this year. The memo assumed civil society attention would be busy elsewhere. Six months of prior work by EFF and EPIC, and one WIRED investigation, ensured it wasn't. Whether that pressure holds the next time Meta quietly ships a feature is a different question and one the company has given no reason to answer confidently.