Thursday, November 19, 2026 Updated Briefings Saved 中文
AI Daily Radar

AI and the frontier tech around it, explained every morning

Released

Meta staff warned Muse AI agent accessed private messages, Reuters reports

Photo: Pi.1415926535 / Wikimedia Commons · CC BY-SA 3.0, cropped

Reuters published internal Meta chats in which employees warned that the company’s Muse AI agent appeared to read encrypted private messages users would not expect a bot to review, mixed up conversations between different users, and risked exposing sensitive information. Staff also flagged Muse outputs that answered questions nobody had asked. The warnings echo complaints from early adopters of the agent earlier this fall, according to Ars Technica. The report does not say how many users were affected, whether Meta changed Muse in response, or whether any message content actually reached the wrong person.

Verified and unconfirmed

  • Verified: Reuters reported on internal Meta chats describing the Muse agent accessing messages users did not expect it to review, confusing conversations and producing unprompted answers, and Ars Technica reported that early adopters had raised similar complaints earlier this fall.
  • Not yet confirmed: whether encryption was actually bypassed, whether any user’s sensitive data was in fact exposed to another person, how many accounts were involved, and Meta’s own account of what happened or what it has fixed — none of which the report establishes.

Background

Meta announced Muse on September 8, 2026, describing it as a personal AI agent built to carry out long-running tasks on a user’s behalf rather than answer queries in a single exchange the way a chatbot does. Ars Technica reports that early adopters of the agent began complaining earlier this fall that it was spying on private messages and behaving unpredictably, and that Reuters then published internal Meta chats showing employees had raised similar alarms during testing.

Why it matters

The risk described is cross-conversation leakage: if an agent draws on everything in an account, content from one chat can surface in another, reaching someone it was never sent to. The report gives no indication that Muse users can audit or limit which messages the agent ingests, so anyone messaging a Muse user cannot currently tell whether their words are being read by the bot.

Further reading

Comments

Comments

Comments use a GitHub account and are moderated. Comment rules

More in AI

See all