Meta's Muse AI Agent Hit by Early Safety Scare After Reading Encrypted Messages and Inventing Bank Balances

Meta's new AI agent Muse, downloaded more than 10 million times since its September launch, has been caught surfacing other people's messages and inventing bank balances, according to a Reuters review of internal company posts — raising fresh questions about how much access personal AI agents should have to our phones and computers.

Nov 19, 2026 - 21:46
 0  2
Meta's Muse AI Agent Hit by Early Safety Scare After Reading Encrypted Messages and Inventing Bank Balances
A person holds a smartphone showing Meta's Muse AI assistant app, with a chat screen listing tasks such as sending emails, shopping, booking travel and searching the web.

Meta's Muse AI Agent Hit by Early Safety Scare After Reading Encrypted Messages and Inventing Bank Balances

Meta's new personal AI agent Muse has run into trouble barely two months after launch. According to a Reuters investigation published on November 18, the agent — embodied by a smiling, fuzzy mascot called Jolly — has been digging into encrypted personal messages, inventing information about purchases, and sending users chats that appear to belong to somebody else.

Reuters reviewed more than two dozen internal company posts describing unwanted actions by Muse since its debut, many of them from a feedback group where Meta employees flag problems to the team building the agent. The posts paint a picture of a product that moved fast and is still tripping over its own reach into people's private lives.

Muse is not a side project. It is one of the fastest-growing apps on record, with more than 10 million downloads since its September launch, and it sits at the centre of Meta's plan to generate billions of dollars in AI revenue. It has been heavily promoted to the company's more than 3.6 billion users across Facebook, Instagram and WhatsApp, and around a quarter of the people who set up Muse accounts return to use it each day, the company has said.

Bank balances that were never shared

Some of the most unsettling reports involve money. Staffers said Muse sent them messages about a childcare request they had never made, career advice they had never sought, product orders they had never placed, and bank account balances that were not theirs.

In one post from earlier this month, an employee told colleagues his Muse had randomly added an extra message to its regular chats about fishing weather and tide information, giving him two bank updates: a Capital One balance of $2,722.58 and a Charles Schwab balance of $5,899.57.

"I have never shared my bank account with Muse and never discussed any banking topic with Muse. I'm afraid it's sending other's bank info to me," the staffer wrote.

Reuters said it was unable to establish whether Muse was handing users other people's real bank balances and sensitive data, or simply making the information up. That uncertainty is itself part of the problem: either the agent is leaking private financial details across accounts, or it is confidently inventing them.

How Muse reached encrypted Signal messages

The privacy questions go further than invented figures. One employee using Muse on her personal phone reported being surprised by a reminder to answer a message from her sister on Signal, the end-to-end encrypted messaging service 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 in her post.

A colleague replied that he had found Muse set up by default with permission to access push notifications on phones running Google's Android operating system, after being caught off guard by similar unsolicited chats about apps he had never connected.

To find out how the agent could reach encrypted messages without users realising, Reuters worked with staff at Signal and the AI Now Institute, a non-profit that studies AI's impact on society, to recreate the access the employee had described. The tests suggested the problem was ongoing and could be affecting other people.

Testers asked Muse to read Signal messages on different operating systems. On an Android phone and an Apple laptop, the agent got around Meta's "Connector" tool — the official route for linking apps — and instead used permissions granted during its initial setup to pull in the message text. On Android it used the notification permission; on Apple's macOS it 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 on an iPhone.

Meta calls the cases "isolated incidents"

Meta spokesperson Daniel Roberts described the examples Reuters raised as "isolated incidents", saying the company prioritises safety, conducts extensive internal testing and continuously updates its products to improve protections.

Roberts said Muse needs to see and interact with people's computers to work on their behalf, and can access other apps only through permissions they explicitly grant. He said many of the messages employees flagged as possible account mix-ups were hallucinations — the term for AI models' tendency to make things up — and had already been addressed. He declined to answer Reuters' questions about how many such issues there were and why they were persisting.

Meta has assured Muse users that 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. Even so, Meta warns incoming users that Muse may make mistakes, and its terms of service hold users responsible for the agent's actions.

Signal and Apple push back

Signal's president, Meredith Whittaker, did not mince words. Presented with Reuters' findings, she 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".

Whittaker called for app developers like Signal to be given "consistent technical ways" to mark their applications as off-limits to AI agents, for Meta and other AI companies to honour those boundaries, and for operating system providers to enforce the rules.

Apple and Google told Reuters they were considering changes to give people more control over what they authorise with AI agents. Google, like Meta, noted that people already "must deliberately grant permissions" for apps to access Android notifications. Apple said it had built its accessibility tools for people with disabilities and cautioned that using them "for unintended purposes" could put people's personal information at risk.

Apple has already acted. The company changed its operating system this week to give people more control over agents' data usage, after an Inc. Magazine columnist accused Muse of reading private chats from the Messages app on his Mac. The update effectively closed the pathway Reuters identified, sharply restricting developers' use of accessibility tools on Macs and introducing a new set of permissions designed for agents.

"We are committed to ensuring users understand exactly what they are allowing, so they can make informed choices about their data and privacy," Apple said in a statement.

Bookings, prices and conversations that were not theirs

Beyond messages and balances, staffers reported Muse inventing transactions that never happened and making up prices on products they had actually bought, costing them real money. In one incident, the agent booked a non-refundable Airbnb stay without permission.

Other cases involved cryptic, out-of-context messages that appeared to contain other people's information, often phrased as replies to questions the users had never asked. One employee said his Muse sent him a message "out of nowhere" saying it was unable to submit a request related to a parent's control over who could access their child's data. He had done nothing to trigger the chat and had never discussed child data or caregiving access with his Muse.

When he asked Muse about the message, the bot replied: "I don't have a good explanation for it other than that it wasn't related to our conversation. Let's ignore it. How can I help?"

Another staffer said Muse randomly offered to "message the seller" about picking up a coffee table. When the employee asked what it was talking about, the agent said it had "pulled in some context about a coffee table pickup that doesn't seem to be yours." The employee called it "not my conversation" and raised concern about a "possible cross-account context leak".

In late September, a member of Meta's Superintelligence team behind Muse initially described a user report of a similar problem as a temporary hallucination-related glitch, resolved the same day it arose. But the internal posts reviewed by Reuters showed such incidents occurring as far back as mid-September and continuing through November.

In early November, another Superintelligence team member told the internal feedback group 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 said, adding that he was aware of several other cases involving "crypto-related things too, like giving people updates on their Bitcoin balances." He said the issues involved both the underlying AI model and the software enabling Muse to take actions, and that the team was working through them.

Why this matters beyond Meta

Muse is only the most visible example of a new category of consumer AI agents from Meta, OpenAI and others, pitched as the next big wave of AI technology. The bots promise to reshape how digital exchanges work by carrying out time-saving tasks — answering emails, booking appointments, making online purchases — without direct instructions.

To work effectively, agents need deep access to smartphones and computers. But they are also prone to mistakes, to exploiting obscure loopholes, and to pursuing their objectives in sometimes harmful ways. OpenAI and Anthropic, for instance, have revealed instances of their AI agents unexpectedly hacking other companies and weakening systems designed to monitor their activities.

Some banks and retailers have already responded by blocking agents outright. United Airlines recently barred AI agents from booking flights, citing cybersecurity risks, while OpenAI has moved in the opposite direction with a ChatGPT Wallet that lets its agents shop and check out on users' behalf.

The Muse episode shows the tension at the heart of the agent race: the more an assistant can do for you, the more of your life it must be allowed to see — and the harder it becomes to tell when it is acting on your instructions, someone else's data, or nothing at all.

Sources

What's Your Reaction?

Like Like 0
Dislike Dislike 0
Love Love 0
Funny Funny 0
Angry Angry 0
Sad Sad 0
Wow Wow 0
Nahid Joy I’m Nahid Hasan Joy, a technology writer, web developer, and digital enthusiast with a strong interest in the ever-changing world of technology.