Meta is keeping your Ray-Ban voice recordings to train its AI

If you are wearing a pair of Ray-Ban Meta smart glasses right now, Meta just changed the deal on what it listens to and keeps.

In an email sent to glasses owners on April 29th, Meta quietly rolled out two major privacy policy updates. The takeaway is simple: the company wants more of your voice data to train its models, and it is taking away your ability to opt out.

Cloud audio storage is no longer optional

Taking a page out of Amazon's playbook, Meta removed the toggle that allowed users to prevent their voice recordings from being stored in the cloud.

Going forward, you cannot disable voice recording storage anymore. You can still go into settings and manually delete recordings whenever you want, but the default is that Meta keeps them.

According to Meta's voice privacy notice, voice transcripts and stored audio are kept for up to a year to help improve products. Even if the system flags an interaction as an accidental wake-up, Meta still holds onto the audio for 90 days before purging it.

The email also stated that Meta AI with camera use is now always enabled on your glasses unless you turn off the hands-free Hey Meta voice command completely.

Meta spokesperson Albert Aydin clarified that photos and videos stored in your phone camera roll are not used to train AI models, even if captured via voice commands. But if you share those photos directly to Meta AI or cloud services, those specific platform rules kick in.

These policy changes started surfacing for some users back in March and officially took effect in the United States on April 29th.

Meta is moving fast across its hardware and software stack. Earlier this month, the glasses gained live translation features. Just yesterday, Meta launched a standalone Meta AI smartphone app to fight ChatGPT and Google Gemini. Reports also suggest Meta is planning a premium pair of Ray-Ban glasses for late 2025 that could jump from the current 299-dollar entry point to around 1,000 dollars.

Why it matters

We think this move shows the real price tag of subsidized AI hardware. Once a gadget gains traction, the data harvesting begins.

For everyday users, the friction is real. If you want hands-free voice commands, your audio is heading to Meta servers for up to twelve months. The only way out is killing hands-free entirely or remembering to manually clear your logs. That is not a user-friendly compromise.

For the AI race, real-world conversational audio captured from first-person wearable mics is pure gold. Meta has the only widely adopted AI glasses on the market, and it clearly wants to feed that sensory stream directly into its models to outpace rivals.

Still, we see serious risk here. Wearables demand a much higher level of consumer trust than phones or browsers. Removing the upfront opt-out switch feels aggressive, and it looks like an open invitation for privacy regulators to step in.