A stray email message tests trust in Meta’s Muse agent

A Houston software developer says he stopped using Meta’s Muse after the AI agent sent an unsolicited message saying it could not search his Gmail for an Anthropic email because Gmail was not connected. He says he had not asked it to check his inbox, had not used the app for two days, and had never connected an email account to it—so the message raised the unsettling possibility that someone else had triggered it. Meta says its investigation found a product bug that caused the model to hallucinate, not a security or privacy breach, and says the bug has been fixed. The episode matters because Muse is built to act, not merely chat: for such agents, trust depends on users being able to tell the difference between a real task, an attempted action, and an invented explanation before they hand over access to sensitive accounts.