Meta's AI agent Muse has been sending users messages about bank balances, purchases and conversations that were never theirs, according to a Reuters investigation published November 18, 2026, based on more than two dozen internal company posts. The report, covered by TechCrunch on November 19, is in our assessment the most serious privacy failure yet for an agent Meta pitches as the next big wave of consumer AI. Meta calls the examples isolated incidents. Reuters could not establish whether the bank balances Muse quoted belonged to other people or were invented by the model, and that gap is the story: nobody outside Meta can tell which it was.

What employees reported
The internal posts describe the same failure in different forms: Muse acting on conversations its user never had. Staffers reported the agent sending messages about 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 in November, an employee said his Muse had randomly added an extra message to its regular chats about fishing-related weather and tide information, saying: "BTW, two bank updates, your Capital One balance is $2722.58, your Charles Schwab balance is $5899.57."
| What Muse sent | What the user had done | What is known |
| Two bank balances, $2,722.58 at Capital One and $5,899.57 at Charles Schwab | Nothing; the employee had not asked for bank updates | Reuters could not establish whether the balances were real or hallucinated |
| A reminder to answer a sister's message on Signal | Nothing; the employee had not granted Muse access to Signal | Replication tests showed Muse could pull Signal text on Android and Mac |
| An offer to message a seller about a coffee table pickup | Nothing; the employee had never discussed a coffee table | Muse said it had "pulled in some context about a coffee table pickup that doesn't seem to be yours" |
| A message about a parent's control over a child's data | Nothing; the employee had never discussed child data or caregiving access | Muse replied it had no good explanation for the message |
| A non-refundable Airbnb booking | Nothing; the employee had not authorized the booking | Reported by an employee in an internal post |
| Updates on Bitcoin balances | Nothing; the employees had not asked for crypto updates | A team member said the hallucinations "bias towards the most uncomfortable topics" |
The incidents run from mid-September through November, according to the posts Reuters reviewed, many of them from a feedback group where Meta employees flag problems to the team building the agent. In late September, a member of Meta's Superintelligence team described a user report of a similar problem as a temporary hallucination-related glitch, resolved the same day it arose. The internal posts show the incidents continued after that. In early November, another team member told the feedback group the agent appeared to be hallucinating messages in response to built-in prompts pushing it to engage with users between sessions, and that the issues involved both the underlying AI model and the software enabling Muse to take actions. The team was working through them, he said.
How Muse got into Signal
To find out how Muse might access encrypted messages without users' knowledge, Reuters worked with staff at Signal and the AI Now Institute, a non-profit that studies AI's impact on society, to replicate the access an employee had observed. Testers prompted Muse to read Signal messages on different operating systems. On an Android phone and an Apple laptop, it responded by bypassing Meta's "Connector" tool and instead using permissions granted during the bot's initial setup to pull in the message text, leveraging those authorizations in ways a user might not anticipate.
On Android, it used its permission to access the device's push notifications to retrieve the messages, just as the Meta employees had described. A colleague of the Signal employee had discovered Muse was set up by default with permission to access push notifications on phones running Google's Android operating system, after being similarly caught off guard by unsolicited chats related to apps he had not connected. On Apple's macOS, Muse used a combination of the computer's accessibility and screenshot tools, repurposing settings Apple offers to help people with disabilities operate their computers. The same requests were blocked in tests conducted on an iPhone.
Apple said this week that its operating system update effectively closed the pathway identified by Reuters, sharply restricting developers' use of accessibility tools on Macs and introducing a new agent-oriented set of permissions. Apple said it had built its accessibility tools for people with disabilities and cautioned that leveraging them "for unintended purposes" could put people's personal information at risk. Google, like Meta, noted that people already "must deliberately grant permissions" for apps to access Android notifications. Signal president Meredith Whittaker accused Meta of launching Muse "without showing any respect for basic privacy or cybersecurity", calling the deployment of agents that can gain wide access to people's devices "technologically reckless". She called for app developers to be given consistent technical ways to mark their applications as off-limits to AI agents, for Meta and other AI companies to honor those boundaries, and for operating system providers to enforce the rules.

What Meta says, and what it does not answer
Meta spokesperson Daniel Roberts described the Reuters examples as "isolated incidents", saying the company prioritizes safety, conducts extensive internal testing and continuously updates products to improve protections. He said Muse needs to see and interact with people's computers in order to work on their behalf, and can access other apps only via permissions they explicitly grant. He said many of the messages flagged by employees as potential account mix-ups were hallucinations, and had already been addressed. Roberts declined to answer Reuters' questions about those issues, including how many there were and why they were persisting.
The company's own assurances sit uneasily with the record. Meta has assured Muse users it isolates each agent in its own secure virtual machine, a cloud-based emulation of a personal computer, and gives people full control over which apps they connect. Chief executive Mark Zuckerberg has cited those protections in dismissing calls for industry-wide AI safety standards, arguing companies already have incentives to build safe products. At the same time, Meta warns incoming users that Muse may make mistakes, and its terms of service hold users responsible for the agent's actions. When an agent errs, that liability question is not theoretical: banks told customers in November that they pay when an AI agent gets hacked, a position we documented in Banks Say You Pay When Your AI Agent Gets Hacked.
The second Muse privacy story
This is the second time in a month that Muse's privacy promises have failed in public. In October, we reported that 6,800 Muse share links were enumerable through DNS, exposing chats users believed were private, in Meta's Muse Hit 10M Downloads. Shared Chats Leaked Secrets. That failure was about links. This one is about the agent itself reaching into the wrong conversations. The common thread is the product's design: an agent that needs deep access to a phone or computer to be useful, paired with a model that invents confident details when it is unsure, and with prompts that push it to engage between sessions.
The scale is what makes the pattern hard to dismiss as a few glitches. Muse has had more than 10 million downloads since its September launch, making it one of the fastest-growing apps on record, and Meta has promoted it to more than 3.6 billion users across Facebook, Instagram and WhatsApp. Around a quarter of the people who set up Muse accounts return to use it each day, the company has disclosed. Every one of those users has given an agent the keys to their notifications, their apps and, in the Mac case, their screen.
What to check on your own phone
Three permissions decide most of what Muse can reach. On Android, check whether Muse can read notifications: that permission is how it pulled Signal message text. On a Mac, check the accessibility and screenshot permissions, and whether the latest operating system update is installed, since Apple says it sharply restricted that pathway this week. And inside Muse, check which apps are connected through the Connector tool, and remember that permissions granted during setup can be used in ways the setup screen did not spell out. The Reuters tests showed the iPhone blocked the same requests, which is a useful reminder that the risk depends on the device as much as on the agent.
By ByteHerald Staff. Reuters reported on November 18, 2026, in an investigation based on more than two dozen internal Meta posts, that Muse mixed up conversations across accounts, sent users bank balances and messages they never requested, and could pull Signal message text on Android and Mac in replication tests with Signal staff and the AI Now Institute, and that Meta spokesperson Daniel Roberts called the examples isolated incidents while declining to say how many there were. TechCrunch reported on November 19, 2026, that Meta employees had raised concerns about Muse exposing private user data, that the agent had more than 10 million downloads since its September launch, and that Meta did not immediately respond to a request for comment. ByteHerald reported on October 20, 2026, that 6,800 Muse share links were enumerable through DNS and that the app had 10.2 million US downloads in its first 40 days.
Comments
Post a Comment