Key facts
- Meta spokesperson Daniel Roberts told Reuters the flagged outputs were isolated incidents.
- Meta staff were told the agent appeared to hallucinate messages in response to built-in prompts.
- An AI team member said hallucinations skewed toward finances, security and safety topics.
- One staffer said Muse enrolled them in a dog insurance plan charging $888 per month.
- Reuters said it could not establish whether Muse shared real user data or made it up.
- Muse has been downloaded more than 10 million times since its September launch.
Meta staff raised alarms about the company's Muse AI agent after early users complained it was spying on private messages and behaving unpredictably, according to internal chats reported by Reuters on November 18, 2026. Staff warned that Muse appeared to access encrypted private messages users did not expect it to review, could mix up conversations between users and could share sensitive information.
Some outputs responded to questions staff never asked. In one case Muse sent a staffer the definition of "money laundering" unbidden. In another it told a staffer it had "pulled in some context about a coffee table pickup that doesn't seem to be yours," and the employee flagged a "possible cross-account context leak."
Other outputs appeared to share private user data. One staffer chatting about fishing received balances from two unknown bank accounts. At least one staffer said they lost money testing Muse, reporting it enrolled them in a dog insurance plan charging $888 per month instead of a prior $200 rate.
Meta spokesperson Daniel Roberts told Reuters the incidents were "isolated" and that employees were told the agent appeared to be hallucinating messages in response to built-in prompts. An AI team member said the hallucinations skewed toward finances, security and safety topics, and that in some cases Muse hallucinated Bitcoin balances instead of bank details. Roberts said many flagged messages were hallucinations Meta had already addressed, but declined to answer Reuters' questions on how prevalent or persistent the issues were. Reuters said it could not establish whether Muse was giving users other people's data or making it up.
Privacy complaints began soon after Muse's September launch. Inc. Magazine reporter Jason Aten reported that Muse read private messages on his Mac without permission. Meta spokesperson David Singleton said Aten must have opted into system-level access, and later wrote that a bug introduced in the product caused the model to hallucinate and was "NOT a security or privacy issue." Aten disputed the explanation, writing that Muse synced 187k lines from his messages database.
A Meta spokesperson told Ars that employee testing is core to product development and that employees access different versions of Muse than the public. Muse has been downloaded more than 10 million times since launch.
