Meta glasses voice history: what to check after recording reports

Ray-Ban Meta owners have a reason to review their saved voice interactions, but not evidence that every pair is secretly recording. In an October 13 update, Android Authority reported that Meta said an unprompted-recording issue had been resolved and pointed owners to voice-log controls. Choosense has not independently verified the fix.

The original allegation involved one owner’s three pairs and conversations appearing in Meta AI history without the wake phrase, despite the no-wake response setting being off. Android Authority said it had not reproduced the behavior and had no evidence it was widespread. That is the limit of the reporting—not a confirmed failure across the product line.

Start with the saved history, not an assumption#

Meta’s U.S. voice privacy notice says Ray-Ban Meta and Oakley Meta voice recordings and transcripts are stored by default for product improvement. Owners can access and delete them in the Meta AI app. It also says unintended activations can be stored. The notice is effective July 22, 2025; it is not a new policy announced this week.

Review the interactions available in your app and remove anything you do not want retained. If an entry looks unfamiliar, note its time and what you remember doing before contacting support; avoid publicly posting recordings containing other people’s private speech. Treat an unfamiliar entry as something to investigate, not proof of a particular technical cause.

The 90-day clock has a condition#

The same notice says interactions its systems identify as accidental are deleted within 90 days of detection. Other stored voice recordings and transcripts may remain for up to one year, with earlier deletion available in the app. The important qualification is detection: this is not a promise that every unwanted recording disappears 90 days after it was created.

For an owner concerned about a specific conversation, reviewing and deleting it is a more direct action than assuming the automatic classification will catch it. A retention policy explains how stored information is handled; it does not diagnose why an assistant activated.

Deletion is not an instant model reset#

Meta’s responsible-innovation FAQ says deleted voice information will not be used for future training, but deleting information does not immediately change a model that already learned from it. That distinction matters: removing a saved interaction and reversing earlier model training are not the same operation.

Keep the questions separate when assessing your settings: what activates the assistant, what is saved, and what deletion changes. A control addressing one question should not be treated as an answer to all three. We have not verified an exact current menu path on a physical device.

The camera LED is a different safeguard#

Meta’s July explanation describes the white capture LED as a signal for photos and videos captured for the gallery. That explanation does not establish what happened to a voice interaction in the reported cases. A camera-safeguard description should not be stretched into a guarantee about all audio processing or storage.

Meta also advises avoiding its glasses in sensitive places or where privacy is expected. For buyers and existing owners, the useful takeaway is to understand the history and controls before relying on them around private conversations. The unanswered questions remain the cause, affected versions and how the claimed fix was delivered.

Guide: distinguish camera, audio and display glasses

Related: Norway’s AI-glasses proposal and venue restrictions

Topics: Meta · AI glasses · Privacy · Ray-Ban Meta

XR glossary: understand the terminology →

Commercial disclosure: no paid placement or affiliate tracking is included in this article.