The accessibility case rests on two categories of users where smartphones consistently fall short: people who cannot see well enough to navigate a screen, and people who cannot use their hands to operate one.

Donald Overton, a blind Army veteran who lost his sight to a blast in Iraq, now uses Meta AI glasses to navigate airports and read dinner menus, replacing what he described as a backpack full of specialized assistive hardware, Meta reported last month. For blind and low-vision users, the glasses offer AI-generated scene description on demand: no stopping, no pulling out a device, no relying on whoever happens to be nearby. The Be My Eyes integration lets users initiate a hands-free video call by saying "Hey Meta, Be My Eyes with [name]," connecting them to a trusted contact or a trained support representative from brands including Tesco, Sony, Amtrak, and Hilton, without touching anything, per Meta's announcement. Real-time captions for incoming calls appear directly on the in-lens display, according to Meta, useful when a phone pressed to your ear is the wrong tool for a busy street.

VA Blind Rehabilitation Centers are using Ray-Ban Meta glasses to support blind and low-vision veterans. Meta worked with the Blinded Veterans Association to develop a training guide covering voice commands, document reading, and call management, per Meta's November 2025 announcement. Institutional adoption of that kind is not proof of clinical efficacy, but it is a different category of signal than a company press release.

For mobility-disabled users, the value is more direct and harder to replicate with software. Noah Currier, a quadriplegic Marine veteran and founder of the Oscar Mike Foundation, used voice commands to take the first photo of his newborn baby the day he brought the glasses home. "I'm a quadriplegic, so my hands don't work. I probably have fewer photos and videos in my phone than anybody else in the world," he said, per Meta. Pressing a shutter button was simply not available to him before. No smartphone software update addresses that.

Both accounts come from Meta's own promotional material. The functional gap they describe is real. Whether the glasses close that gap reliably, at scale, against dedicated assistive hardware, remains an open question without independent data.

None of the accessibility functions described above run locally. Every AI feature, including scene description, voice commands, and call management, requires footage or audio to be sent to Meta's servers. On-device processing is not available for any of them, EFF confirmed earlier this year. When a blind user says "Hey Meta, what am I looking at?", that footage travels to Meta's infrastructure before an answer comes back.

Photos and videos sync automatically by default into the Meta AI app, which is required to set up the glasses, though EFF noted that users can operate the glasses without the app, since downloading footage without it is cumbersome rather than impossible. Audio from AI interactions is saved by default unless manually deleted after each session. Some footage is used to train Meta's AI models and may be reviewed by human contractors. An investigation by Swedish journalists found workers annotating camera footage that included nudity, bathroom use, and other sensitive content. Meta confirmed the practice to the BBC as consistent with its terms of service, as reported by EFF. Employees can potentially access that footage, and it may be shared with law enforcement.

Consider what that means in practice. A blind user asking the glasses to describe a hospital waiting room, a domestic violence shelter, or a private medical appointment captures footage of people who gave no consent to being recorded. They almost certainly don't know it happened. The glasses are designed to look like ordinary eyewear, and multiple reviewers noted that friends didn't notice the embedded cameras. The indicator light that activates during video recording can be disabled through cheap hacks, EFF noted. These are not far-fetched scenarios. They are exactly the environments disabled users depend on these glasses to navigate.

Two groups end up with very different stakes. Disabled users gain meaningful autonomy: navigation, communication, documentation. People around them typically get little notice and no practical opt-out on whether their face or voice ends up in Meta's training pipeline or surfaces in a law enforcement request.

Earlier this month, Wired and EFF confirmed that Meta had shipped facial recognition code inside the Meta AI app to millions of devices. EFF's Threat Lab verified the code's presence through static analysis. The code was present and active in the application, though not yet exposed to consumers, EFF reported. A separate researcher demonstrated that after manually adding a faceprint through debug mode, the glasses would subsequently detect and identify that face when it entered view.

The system stored faceprints as sequences of 2,048 numbers representing the geometry of a person's facial features. When activated, it would convert every new face in the glasses' sightlines into a numerical representation and compare it against a stored database, according to EFF. The code was not in consumers' hands as a usable feature, but it was running on shipping hardware.

Following Wired's reporting and the public attention it drew, Meta removed the code in an app update that EFF confirmed days later.

Meta's history makes the pattern harder to dismiss as an isolated mistake. The company previously paid $650 million to settle a class-action suit under Illinois's Biometric Information Privacy Act for collecting facial data from photos without user consent, a feature it subsequently shut down, per EFF. Internal documents showed the company had discussed timing a facial recognition launch to coincide with periods of political distraction, when civil society groups "would have their resources focused on other concerns," EFF reported.

The facial recognition episode doesn't erase the accessibility story. It clarifies who is telling it. Disabled users who rely on these glasses to move through the world are, by definition, among the least positioned to audit what the device is doing in the background, modify weak default settings, or choose a privacy-respecting alternative that does not yet exist.

The accessibility benefits are real. They are also almost entirely documented through Meta's own promotional material: no independent adoption figures, no published reliability comparisons against dedicated assistive tools, no data on how often scene description fails in conditions that matter.

The privacy architecture is not a separable concern. Every feature that helps a user runs through the same cloud infrastructure that, earlier this month, contained active facial recognition code confirmed by independent analysis, EFF, June 2026. The question critics have raised, whether accessibility features could work with meaningfully different privacy protections including on-device processing, opt-in cloud features, or automatic face redaction in uploaded footage, is not hypothetical. Google began redacting faces in Street View years ago following sustained pressure from privacy advocates including EFF, EFF noted earlier this year. The technical path exists. Meta has not taken it.

What Meta has not provided is any public commitment to on-device processing for accessibility features, any framework for bystander privacy, or any explanation of why facial recognition code was present in a shipping app before it was discovered and removed. Until those questions have answers beyond a removal prompted by a news cycle, the most compelling use case for Meta's smart glasses is also the one with the most to lose if the underlying architecture stays as-is. The people who need these glasses most deserve hardware that works for them, not hardware that works for them while quietly working for everyone else too.