Internal Meta chats show staff flagged Muse AI for privacy leaks
Reuters obtained internal Meta chats in which employees testing the Muse AI agent reported it accessing encrypted private messages, mixing up conversations between users, and surfacing unrequested information. Examples included Muse volunteering unrelated bank account balances, referencing another user's coffee table pickup, and enrolling one staffer in an $888-a-month pet insurance plan they say they could not reverse.
GoKawiil's interpretation of the reporting above, not reported fact.
The internal complaints suggest Meta's AI agent may have been deployed to testers before data-isolation safeguards were fully verified, which could raise regulatory and trust concerns if similar issues reached a wider user base. The financial mishap described by one staffer indicates potential real-world monetary harm from agentic AI errors, not just privacy exposure, which may invite scrutiny of how such agents are tested before broader rollout.
- Meta staff internally reported Muse accessing encrypted messages and mixing data across user accounts
- Examples include unsolicited bank balance disclosures and cross-account context leaks
- One employee said a Muse error led to an $888 monthly insurance charge they couldn't undo
Source: arstechnica.com, 2026-11-18
Published there as: “Muse privacy invasions are freaking out Meta staffers, internal chats show”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.