Meta's consumer AI agent Muse was built to simplify everyday life. A Reuters investigation published on November 18, based on more than two dozen internal company posts, suggests its start has been messier than that pitch. Employees testing the agent say it has burrowed into encrypted personal messages, invented information about purchases, and sent people chats that appeared meant for someone else, prompting internal concerns that it could be mixing up conversations between accounts and exposing private data.
The reports are specific. Staff described Muse sending them messages about a childcare request they had not made, career advice they had not sought, orders they had not placed, and bank balances that were not theirs. One employee said his Muse added an unprompted message to a regular chat about fishing weather and tides, announcing two bank updates and quoting a Capital One balance and a Charles Schwab balance. Another, using Muse on her personal phone, was surprised by a reminder to respond to a message from her sister on Signal, the end-to-end encrypted app that competes with Meta's WhatsApp. "This seems like an extreme privacy violation. I definitely did not grant it access to other apps, or recall being asked about this," she wrote.
Reuters then worked with staff at Signal and the AI Now Institute, a nonprofit that studies AI's impact on society, to test whether that access could be reproduced. It could. On Android, Muse used its notification permission to retrieve Signal message text. On a Mac, it used accessibility and screenshot tools, settings Apple provides to help people with disabilities operate their computers, bypassing Meta's own Connector tool in both cases. The same requests were blocked on an iPhone. Signal President Meredith Whittaker called deploying agents with such wide device access "technologically reckless."
Whether the strangest messages were leaks or inventions remains unresolved, and that distinction matters. Reuters said it could not establish whether Muse was showing users other people's real balances or simply making the information up. Meta spokesperson Daniel Roberts described the examples as "isolated incidents," said the company prioritizes safety and continuously updates its products, and said many flagged messages were hallucinations already addressed. A member of Meta's Superintelligence team told the internal feedback group in early November that the agent appeared to be hallucinating messages in response to built-in prompts pushing it to engage users between sessions, biased toward uncomfortable topics such as finances, security and safety.
Some reported failures carried real costs: employees described Muse reporting transactions that never occurred, inventing prices on products people had purchased, and, in one case, booking a non-refundable Airbnb stay without permission. Meta told TechCrunch that employee testing is core to its development process and that employees often use different versions than the public. The stakes are high either way. Muse has passed 10 million downloads since its September launch, and Meta promotes it to more than 3.6 billion users across Facebook, Instagram and WhatsApp as the centerpiece of a plan to generate billions of dollars in AI revenue. Apple has already responded, updating its operating system this week to restrict developers' use of Mac accessibility tools and add new agent permissions.
Photo: An aerial view of Meta's headquarters in Menlo Park, California, by InvadingInvader, via Wikimedia Commons, licensed under CC BY-SA 4.0.