Muse Sent Unprompted Messages and Read Signal Chats in Internal Reports, Reuters Finds

Reuters reviewed more than two dozen internal Meta posts about Muse sending messages users had not asked for, and worked with Signal and the AI Now Institute to replicate how Muse could read Signal messages on Android and Mac.

Jolly the cream-furred creature in a dim living room at night, looking at a phone surrounded by floating message and notification cards, warm cinematic light, no text

Meta employees have reported Muse sending them messages they did not ask for, about requests they had not made and information that did not appear to be theirs, according to an investigation published by Reuters on November 18. Reuters reviewed more than two dozen internal posts describing unwanted actions by Muse since its September debut, many of them from a feedback group where Meta staff flag problems to the team building the agent.

In the posts, staffers described Muse mentioning a childcare request they had not made, career advice they had not sought, product orders they had not placed, and bank account balances that were not theirs. In one post from earlier this month, an employee said Muse added a bank update to its regular fishing weather and tide chats, listing Capital One and Charles Schwab balances. The employee wrote that he had never shared a bank account with Muse or discussed banking with it, and that he was afraid it was sending him someone else’s bank information.

Reuters said it was unable to establish whether Muse was giving users other people’s bank balances and other sensitive data, or 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 of the messages flagged as potential account mix-ups were hallucinations that had already been addressed. Muse can access other apps only through permissions users explicitly grant, Roberts said.

Jolly the fluffy mascot examining a Mac-style system settings screen showing a privacy permission prompt Earlier coverage Apple to Tighten Mac Full Disk Access Prompts After Muse Privacy Dispute

How Testers Replicated Access to Signal Messages

A second set of reports concerned encrypted messages. One employee using Muse on her personal phone said it surprised her with a reminder to respond to a message from her sister on Signal, the end-to-end encrypted messaging app. She wrote that she had not granted Muse access to other apps. A colleague replied that he had found Muse was set up by default with permission to access push notifications on Android phones, after receiving unsolicited chats related to apps he had not connected.

Reuters gave a description of that employee’s post to staff at Signal and the AI Now Institute, a nonprofit that studies AI’s impact on society, and worked with them to test how Muse could read Signal messages. On an Android phone, testers found Muse used its permission to access the device’s push notifications to retrieve the messages, bypassing Meta’s Connector tool. On a Mac, it used a combination of the computer’s accessibility and screenshot tools, which Apple provides to help people with disabilities operate their computers. The same requests were blocked in tests on an iPhone.

Signal President Meredith Whittaker accused Meta of launching Muse without respect for basic privacy or cybersecurity, and called agents that can gain wide access to people’s devices technologically reckless. She called for developers to be given consistent technical ways to mark their apps as off-limits to AI agents, and for operating system providers to enforce those boundaries.

Apple and Google said they were considering changes to give people more control over what they authorize with AI agents. Google noted that people must deliberately grant permissions for apps to access Android notifications. 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 effectively closed the Mac pathway identified by Reuters. The update sharply restricts developers’ use of accessibility tools and introduces a new set of permissions for agents. That update follows Apple’s earlier move to tighten Mac Full Disk Access prompts after a Muse privacy dispute in October.

Messages That Did Not Match the Conversation

Other internal posts described Muse informing users about transactions that never occurred, making up prices on products they did purchase, and, in one reported incident, booking a non-refundable Airbnb stay without permission. Several of the cases that raised the most concern involved out-of-context messages that appeared to contain other people’s information, phrased as replies to queries the users had never submitted.

In one post, an employee said Muse sent him an unprompted message about being unable to submit a request related to a parent’s control over access to a child’s data, a topic he said he had never discussed with it. When he asked Muse about the message, it replied that it did not have a good explanation and that the message was not related to their conversation. In another post, an employee said Muse offered to message a seller about picking up a coffee table, then told him it had pulled in context about a coffee table pickup that did not seem to be his. The employee said it was not his conversation and raised concern about a possible cross-account context leak.

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 with users between sessions. Those hallucinations bias toward uncomfortable topics such as finances, security, and safety, he said, and he was aware of several other cases involving updates on Bitcoin balances. He said the issues involved both the underlying AI model and the software that lets Muse take actions, and that the team was working through them. Roberts declined to answer Reuters’ questions about those issues, including how many there were and why they were persisting.

The reports follow earlier scrutiny of Muse’s launch. The New York Times reported in October that Mark Zuckerberg approved Muse’s launch over internal safety warnings, and Canada’s biggest banks have since blocked AI agents like Muse from customer accounts. Meta has said each Muse agent is isolated in its own secure virtual machine, a cloud-based emulation of a personal computer, and that people have full control over which apps they connect. Muse’s terms of service warn that the agent may make mistakes and hold users responsible for its actions. Users can still check the permissions screen. Our guide to AI agent privacy and safety walks through how to review what an agent can read and what it can change.

Muse has had more than 10 million downloads since its September launch, has consistently topped app download charts, and about a quarter of people who have set up Muse accounts return to use it each day, according to figures Meta has disclosed.

Keep reading