Key facts
- Meta's Muse AI agent sent an unsolicited message about checking email.
- The user had not prompted the agent to check email.
- Meta stated the incident was a bug causing the model to hallucinate.
- The bug has reportedly been fixed.
Ugo Nwune, a 37-year-old software developer, decided to try Meta's Muse AI agent after downloading it about four weeks ago. He used it once to assist with reaching out to a prospective investor. Three days later, Nwune received an unexpected message from the Muse agent stating it could not check his email due to a lack of connection. Confused, Nwune interacted with the agent, which indicated it could not determine the cause of the message and offered to report it as a potential integrity issue. Nwune reported the incident to Meta, and a Meta employee suggested it was a hallucination. Meta later confirmed the issue was a bug causing the model to hallucinate, not a security or privacy breach, and that the bug has been fixed.