User reports raise questions about activation and stored conversations; the scope and cause remain unclear.
Ars Technica reports that some Ray-Ban Meta owners found unexpected audio recordings and transcripts despite not deliberately activating the assistant. Meta said the issue had been resolved. The account does not establish how many devices were affected or confirm what caused the behavior. User reports about a firmware update remain unconfirmed.
In plain English
“Firmware” is the software built into a device. A “wake phrase” like “Hey, Meta” is the words that tell an assistant to start listening. A “transcript” is the written text of recorded speech.
Why it matters
Why it’s a red flag The risk is not just the bug: wearable assistants that record by mistake capture other people’s conversations, and the maker has given no cause or count.
A wearable assistant changes the privacy problem because it sits in shared physical spaces. A mistaken recording can involve people who do not own the device, did not configure it and may not know an interaction occurred. That makes clear activation behavior an essential part of the product experience.
The distinction between a reported fix and an explanation is important. Resolving an issue addresses current behavior. Explaining its cause, scope and the status of any stored recordings helps affected people understand what happened. The available account leaves several of those questions open. It does not establish that every pair of glasses was affected.
For an owner, reviewing available voice-history controls can help identify unexpected entries. That is different from assuming a setting was responsible for the bug. The reporting does not confirm a cause, and changing an unrelated option would not prove the underlying issue is resolved.
For product teams, the lesson is about observability as well as consent. People need a recognizable signal when recording begins, an accessible record of stored interactions and a clear way to stop or remove them. These are design considerations, not findings about which controls failed in this incident.
The next useful information would be a fuller account from Meta: what triggered the behavior, which devices or versions were affected and how the fix was verified. Until then, the company statement and user reports should remain separately attributed.
The owner and a nearby conversation participant may have different information about the same event. A useful response needs to address both perspectives: what the device displayed and what it actually retained. The available reporting does not provide that complete reconstruction.