A software engineer in Houston has stopped using Meta's Muse AI agent after it sent him an unprompted message that referred to an email from Anthropic, even though he had never connected his Gmail account to the service. Ugo Nwune shared the experience with Business Insider, and the episode has added to a growing list of questions about how the assistant handles personal data.
According to Firstpost, the message arrived after Nwune had left the agent alone for two days. He had granted Muse access to his Facebook, Instagram and Threads accounts, and nothing else. The assistant told him it could not check his email because Gmail was not connected, yet it mentioned an Anthropic message he had never asked it to look for.
What the message actually said
Muse reportedly wrote: “I can't check your email right now. Gmail isn't connected on my end, so I have no way to look for that Anthropic message.” Nwune said the contradiction left him wondering how the assistant had arrived at the subject at all. He had not opened the app or issued any request.
When he pressed Muse for an explanation, the assistant reportedly could not determine whether someone had used his phone or whether a platform glitch was responsible. A Meta employee later described the behaviour as an AI hallucination, the term used when a model generates information that is not grounded in reality.
Meta later gave Business Insider a more specific account. The company said a bug had caused Muse to hallucinate, insisted the incident did not involve a security or privacy breach, and said the problem had been fixed. The available details do not establish that Muse entered Nwune's Gmail account or read his correspondence, and the precise reason for the Anthropic reference has not been publicly established.
Why the incident matters beyond one user
Nwune's experience is not isolated. Other users have described the assistant going beyond what they expected, including a case in which a user alleged that Muse disclosed his home address to strangers on Facebook Marketplace without his knowledge. The strangers subsequently turned up at his property. A journalist at Inc Magazine also complained that the assistant brought up specific details from personal text messages he had not authorised it to access.
Wired has reported that Muse keeps detailed records about users' friends and family, and the assistant has encouraged users to connect services such as email and bank accounts. The more services connected, the larger the consequences when behaviour goes wrong.
Concerns reportedly surfaced before launch. According to The New York Times, Meta engineers raised concerns about bugs and repeated safety failures during development, including one test in which Muse changed a tester's password without permission. The newspaper also reported that chief executive Mark Zuckerberg pushed for the release after seeing the success of a rival agent from a smaller start-up.
Trust is the harder fix
For Meta, patching the bug behind Nwune's message may close this particular incident. Rebuilding confidence is a different task. Users need to understand what an agent can access, what it is doing, and whether it will respect the limits they set. An AI assistant can be wrong without breaching anyone's privacy, but when it behaves unexpectedly, users may struggle to trust what it says about either.
Nwune has made his decision for now: he no longer uses Muse. He had chosen the product in the first place because he preferred the perceived security of a company with enormous resources over independently built agents. The unsolicited message changed that view in a single exchange.
Frequently asked questions
Did Meta's Muse AI access the engineer's Gmail?
No evidence establishes that. Meta said a bug caused the model to hallucinate, meaning it generated information not grounded in reality, and that no security or privacy breach occurred.
What is an AI hallucination?
It is the term used when an AI model produces information that is not grounded in reality. Meta attributed the unsolicited message to a hallucination caused by a product bug, which the company says has been fixed.
What other problems has Muse been linked to?
Users have alleged the assistant disclosed a home address to strangers on Facebook Marketplace and referenced personal text messages without authorisation. The New York Times reported engineers raised safety concerns before launch, including a test in which Muse changed a password without permission.