Meta says it fixed Ray-Ban glasses that recorded conversations without the wake phrase

Kindly-Audience-137, a Reddit user, said the Meta AI app "saves audio recordings and text transcripts of these conversations in my conversation history." ^[9] The user said the "Respond without Hey Meta" setting was turned off. ^[8] "I believe this deserves a serious technical investigation. Customers should not have to wonder whether their private conversations are being captured when they have not intentionally activated their glasses' AI assistant," the user said. ^[10]

The user reports it on three pairs, each with the setting off, and a transcript showed up in the app anyway. ^{[8] [9]}

Meta's reply to Android Authority read: "This issue has been resolved. People can manage, review, and delete voice logs of their interactions with Meta AI via settings at any time." ^[3] The second sentence sends users to their own settings to review and delete logs. Meta did not say what caused the recordings. The user said they "first noticed this behavior after a recent firmware update," and Meta has not confirmed that a faulty update was behind it. ^[11]

Meta's earlier privacy fix was a mandatory update that disables the camera if someone tampers with the LED privacy light, 9to5Google reported. ^[5] That update targets the camera, and the bug here was audio only. ^[4] Privacy advocates had already raised concerns about the glasses' inconspicuous look, Ars Technica reported. ^[7]

In our view the stored log is where the liability sits, because it is a record of a conversation the wearer did not start. Anyone who allows these glasses on site, or ships a wearable with a wake-word assistant, can sort each device on two questions: whether capture can start without an explicit act by the wearer, and whether the capture is kept as a transcript in an account. If a device can misfire but keeps nothing, it is a nuisance; if it keeps transcripts only after an explicit act, it is a retention-policy question.

Where both are true, as in this report, we would test before rollout. Leave a pair on a desk with the setting off for a day, then read the conversation history. The test costs a spare pair and an hour of someone's time, and a clean result shows only that this pair stayed quiet that day.