Meta fixes glasses that recorded conversations without the activation command

Meta confirmed and fixed a flaw in the Ray-Ban Meta Glasses. The device recorded everyday conversations even without activation by the phrase "Hey Meta."

The case gained traction on Reddit. It then reached the company, which reported the fix without detailing the cause.

How the problem appeared to Meta users

The main interaction with these glasses happens through voice, in a way similar to Alexa and Google Nest.

In theory, Meta's device should only register and process the interaction after the activation command.

In practice, the microphone turned on in an apparently random way. Everyday conversations ended up being recorded without any invocation of the assistant.

One user's account sums up the scenario. However, the glasses recorded while he was simply talking with another person.

Afterward, the conversation appeared listed in the account's transcript history.

The Meta technical detail that draws attention in this Meta case

One point deserves attention for those who work with voice. The assistant received the audio, yet remained inactive.

In other words, the system apparently recognized that it was not a command addressed to it.

Even so, the material still went to transcription. Moreover, capture and interpretation, therefore, operated in separate layers.

Another element reinforces the flaw. According to the account, the LED that signals recording lit up during everyday conversations.

Three clues that point to the firmware

The first sign is in the settings. The option that allows interaction without the wake phrase was turned off.

The second sign comes from the history. The recordings appeared logged in the user's account.

The third sign is the most revealing. However, the behavior started after a recent firmware update.

In the comments on the post, several users reported a similar situation. However, the number of affected devices remains unconfirmed.

The support recommendation that disabled the product

Here the story gets more uncomfortable. The user contacted the company's support and received a list of things to disable.

The guidance included revoking microphone and camera permissions. It also asked the user to turn off Usage Detection, transcript history, and automatic update installation.

In practice, this combination disables core functions of the product.

The user's complaint followed this line. He also questioned the need to disable the main functionality of an expensive AI product just to keep private conversations from showing up in the history.

What Meta said about the fix

Questioned by the site Android Authority, the company avoided detailing the cause of the unexpected recordings.

It only stated that the problem should no longer occur.

Consequently, three public gaps remain. It's still unknown what the root cause was, how many devices were affected, and what happened to the audio already transcribed.

What this Meta episode teaches voice developers

First, treat wake word detection and the recording pipeline as independent layers. The flaw here shows one operating without the other.

Second, log the state of the activation flag alongside every capture. This log also makes it easier to audit when behavior deviates from what's expected.

Third, tie the visual indicator to the actual state of the microphone. An LED that lights up on its own becomes the user's first warning sign.

Finally, think about the support script. Instructing the customer to shut down the entire product rarely works as a response to a privacy incident.

Follow our profile on Instagram!