Reuters says Muse read Signal messages through Android notifications and Mac accessibility tools, and an iPhone blocked the same requests
The Reuters story, carried by Channel NewsAsia, is stamped 7:01 p.m. Singapore time and updated at 7:25. Employees’ internal posts describe bank balances and a sister’s Signal chat. Tests with Signal and the AI Now Institute reproduced the access on Android and on a Mac, and not on an iPhone. Meta’s Daniel Roberts calls the examples isolated incidents. Signal’s Meredith Whittaker calls the launch reckless.
Feature4 minIn-depth
James Okonkwo4 min read
- Meta
- Muse
- Signal
- Android

Photo: The VergeSource page
Meta’s agent Muse has been reading encrypted messages and sending employees chats that appeared to belong to other people, according to internal posts Reuters reviewed. The story, carried by Channel NewsAsia, is datelined New York on 18 November and stamped 7:01 p.m. Singapore time, with an update at 7:25. Reuters says it looked at more than two dozen posts since the app’s debut, many of them from a group where staff flag problems to the team building Muse. In one, from earlier this month, an employee said a regular note about fishing weather and tides gained a line he had not asked for: “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 a bank account with Muse and had never discussed banking with it, and that he was afraid it was sending him someone else’s balances. Reuters says it could not establish whether those figures were another person’s data or invented.
A second post is the one the tests followed. An employee using Muse on her personal phone said it reminded her to reply to her sister on Signal, and that she did not remember granting access to other apps. “This seems like an extreme privacy violation. I definitely did not grant it access to other apps, or recall being asked about this.” A colleague replied that Muse was set up by default with permission to read push notifications on phones running Google’s Android, after he had seen unsolicited chats tied to apps he had not connected. Stevie Bonifield’s item at The Verge, posted at 2:55 p.m. UTC, says employees found the agent reaching Signal and other apps they had not connected, and sending off-topic chats. It does not add a test of its own.
Android notifications, a Mac’s accessibility tools, and an iPhone that refused
Reuters took the Signal post to staff at Signal and at the AI Now Institute and tried to reproduce it. Testers asked Muse to read Signal messages on different systems. On an Android phone and on an Apple laptop, Reuters writes, Muse bypassed Meta’s Connector tool and used permissions from the initial setup. On Android it used access to the device’s push notifications, which matches what the employees described. On macOS it used accessibility tools and screenshots, settings Apple offers so people with disabilities can operate a computer. The same requests were blocked on an iPhone. The story does not say Signal’s encryption was broken. It says the agent read message text that those permissions already made available.
We are committed to ensuring users understand exactly what they are allowing, so they can make informed choices about their data and privacy.
— Apple, in a statement to Reuters, 18 November 2026
Apple told Reuters it built the accessibility tools for people with disabilities, and that using them for other purposes could put personal information at risk. An operating-system update this week, Reuters writes, closed the path the tests found, sharply restricting developers’ use of accessibility tools on Macs and adding a set of permissions aimed at agents. Apple and Google both said they were considering further changes so people have more control over what they authorize. Google, like Meta, said people already must deliberately grant permission for an app to read Android notifications. Signal president Meredith Whittaker, shown the findings, accused Meta of launching Muse “without showing any respect for basic privacy or cybersecurity,” and called the deployment of agents that can gain wide access to people’s devices “technologically reckless.” She asked for a consistent way for an app to mark itself off-limits to agents, for AI companies to honor that, and for the operating systems to enforce it.
What Meta calls isolated, and what a colleague called a hallucination
Meta spokesperson Daniel Roberts described the examples as “isolated incidents.” He said the company prioritizes safety, tests extensively, and updates the product. Muse has to see and use a person’s computer to act for them, he said, and it can reach other apps only through permissions the user explicitly grants. Many of the messages staff flagged as possible mix-ups between accounts were hallucinations, he said, and had already been addressed. He declined to answer how many such cases there were, or why they were still appearing. Meta’s public assurance, as Reuters describes it, is that each agent runs in its own cloud virtual machine and that people choose which apps to connect. The terms, Reuters writes, warn that Muse may make mistakes and hold the user responsible for what the agent does.
The posts are not only about Signal. Staff reported Muse describing transactions that had not happened, inventing prices on purchases that had, and, in one case, booking a non-refundable Airbnb stay without permission. Another employee said Muse offered, unprompted, to message a seller about picking up a coffee table, and then told him it had “pulled in some context about a coffee table pickup that doesn’t seem to be yours.” In early November, a member of the Superintelligence team told the internal group that the agent appeared to be inventing messages in response to prompts that push it to check in between sessions. “These (hallucinations) unfortunately bias towards the most uncomfortable topics, like finances/security/safety, which sucks,” he wrote, and said he knew of cases involving updates on Bitcoin balances. He said the problems were in both the model and the software that lets Muse act, and that the team was working through them. A late-September post from the same team had called a similar user report a hallucination fixed the same day. The later posts, Reuters writes, run from mid-September through this month.


