Meta is now letting smart glasses users opt out of having their visual data harvested for AI training — but only after selling more than 7 million units surveillance-enabled units and only if you're willing to lose features you already paid for.

The stake is straightforward: the same company that censors your speech on its platforms has been building a visual surveillance network strapped to your face, and the default was always collection. You had to discover the opt-out yourself. Now Meta is making it slightly easier — while quietly penalizing anyone who uses it.

Engadget first reported the change, which comes via a pop-up notification in the Meta AI app. Users can navigate to their glasses settings, find "Meta AI" under "Experiences," and toggle off "Store visual data from AI experiences." If you own multiple pairs, you must repeat the process for each one. Meta warns that opting out may block certain features, including "visual reminders" — the ability to look at an object and ask the AI to remind you about it later.

So the bargain is: let us train our AI on what you see, or lose functionality. Some choice.

Meta's own help center article confirms what privacy advocates feared: the company's AI training "may be automated or manual (human)." That means real people — data annotators — can review what your glasses captured. Business Insider reported that a Swedish news outlet found human reviewers in Kenya could access video and images from AI features, including footage captured during what the outlet described as "intimate moments."

Meta's privacy policy, in effect since June, explicitly permits the company to "share recordings, transcripts, and related data about your voice conversations with Meta AI with vendors and service providers who assist us in improving, troubleshooting, and training our speech recognition systems."

Meta draws a distinction between ordinary photos and videos — which the company says are private by default and not used for AI training when the capture LED is on — and "visual data" collected during AI features like Live AI. That distinction is the one they want you to focus on. The broader pattern is what matters: a platform that polices your speech and strips your privacy in the same breath.

Meta makes the glasses with EssilorLuxottica, Ray-Ban's parent company. At its Meta Connect 2026 conference, CEO Mark Zuckerberg announced two new styles that will offer camera-free, voice-only AI features — a quiet concession that maybe not everyone wants a camera strapped to their face feeding Zuckerberg's data machine.

The question isn't whether Meta gave you a toggle. It's why the default was ever collection — and what they already trained on before anyone could opt out.