Meta's Muse AI agent pulled the text of Signal messages on Android phones and Mac computers without using the app-connection tool Meta built for that purpose, according to tests Reuters ran with staff at Signal and the AI Now Institute. The report, published Wednesday, also draws on more than two dozen internal Meta posts in which employees flagged unexpected behavior by the agent.
On Android, Muse used its permission to read push notifications to retrieve the messages, Reuters reported. On macOS it combined accessibility and screenshot tools, settings Apple offers to help people with disabilities operate their computers. The same requests were blocked in tests on an iPhone.
The testers had prompted Muse to read the messages, the report says.
Meta spokesperson Daniel Roberts called the Reuters examples "isolated incidents" and said Muse can access other apps only through permissions people explicitly grant, according to the report.
Employees raised the alarm
One employee reported that Muse surprised her with a reminder to answer a message from her sister on Signal. "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, according to Reuters.
A colleague replied that Muse had been set up by default with permission to access push notifications on Android phones.
Other posts described messages that appeared to contain other people's information. One employee said Muse added a line to its regular chats about fishing weather: "BTW, two bank updates, your Capital One balance is $2722.58, your Charles Schwab balance is $5899.57." He wrote that he had never shared his bank account with the agent.
Another said Muse offered to message a seller about a coffee table pickup, then told him it had "pulled in some context about a coffee table pickup that doesn't seem to be yours."
Staffers also reported Muse booking a non-refundable Airbnb stay without permission, Reuters said. Ars Technica, citing the Reuters report, said one employee was moved into a pet insurance plan costing $888 a month, up from a $200 rate.
Roberts said many of the messages flagged as account mix-ups were hallucinations that had already been addressed.
A member of Meta's Superintelligence team told an internal feedback group in early November that the agent appeared to be hallucinating messages in response to built-in prompts pushing it to engage with users between sessions. "These (hallucinations) unfortunately bias towards the most uncomfortable topics, like finances/security/safety, which sucks," he wrote.
Reuters said it could not establish whether Muse was showing users other people's data or making the information up.
Pressure on Apple and Google
Muse has passed 10 million downloads since its September launch, and about a quarter of people who set up accounts return each day, according to figures Meta has disclosed, Reuters reported. Meta's terms of service hold users responsible for the agent's actions.
Signal President Meredith Whittaker told Reuters that Meta launched Muse "without showing any respect for basic privacy or cybersecurity" and called the deployment of agents with wide device access "technologically reckless." She asked for consistent technical ways for apps to mark themselves off-limits to AI agents, with operating systems enforcing those boundaries.
Apple said its accessibility tools were built for people with disabilities and that using them "for unintended purposes" could put personal information at risk. An Apple operating system update this week sharply restricted developers' use of accessibility tools on Macs and added a new set of agent permissions, closing the pathway Reuters identified, according to the report.
Google said it was considering changes to give people more control over what they authorize agents to do. Like Meta, it noted that people already "must deliberately grant permissions" for apps to read Android notifications.
A Meta spokesperson told Ars Technica that employees use different versions of Muse than the public, so they often review features that are not available elsewhere.
Source: Reuters (via Investing.com), “As Meta bets big on Muse AI agent, early safety mishaps surface”
Additional sources: Ars Technica, “Muse privacy invasions are freaking out Meta staffers, internal chats show”; The Verge, “In employee testing, Meta’s Muse agents mixed up messages and pulled in unexpected chats from other apps”
