# Meta Muse agent mix-ups — staff got bank balances not theirs, Signal read

> Meta employees say the Muse agent showed them data that was not theirs and acted without asking.

- Org: Reuters
- Date: 2026-11-18
- Category: security
- Source: https://techcrunch.com/2026/11/19/metas-ai-agent-muse-is-mixing-up-conversations-and-exposing-private-data-report-says/
- Canonical page: https://ai-tldr.dev/releases/meta-muse-agent-privacy-mishaps-reuters/

## In short

A Reuters investigation of 24+ internal Meta posts found that the Muse AI agent sent staff bank balances that were not theirs, made bookings no one asked for and read Signal messages on Android and Mac. Meta calls the cases isolated.

## What it is

Reuters reports that Meta's Muse AI agent sent employees messages that seemed meant for other people: childcare requests, career advice, product orders and bank balances they had never shared. In one chat about weather and tides, Muse added balances from two banks that the employee had never shared. Staff also described a non-refundable Airbnb booking made without permission and invented prices on real purchases.

## How it works

Reuters, the AI Now Institute and Signal staff tested whether Muse could read Signal messages. On Android it got the text through push-notification access, which Muse had by default; on a Mac it used accessibility and screenshot tools. The same requests were blocked on an iPhone. Reuters found no sign that Signal's encryption was broken, and could not tell whether the bank figures were real data from other accounts or made up by the model.

## Why it matters

Agents that act on a phone or computer inherit every permission the operating system gives them, so a model error can turn into a real booking or a privacy leak. Signal president Meredith Whittaker called Meta's approach "technologically reckless" and asked for a standard way to mark apps off-limits to AI agents. Apple tightened Mac accessibility permissions for agents this week, and Apple and Google say they are considering more agent permission controls.

## Who it's for

Muse users, Android and Mac users of AI agents, and teams building agent permission models

## Quick facts

- Product: Meta Muse AI agent (launched September 2026)
- Downloads: More than 10 million
- Evidence: Reuters review of more than two dozen internal Meta posts
- Period: Mid-September into November 2026
- Signal access: Read on Android and Mac in tests; blocked on iPhone
- Meta's position: Isolated incidents; many were hallucinations, already addressed

## FAQ

### Was Meta Muse really showing other people's bank balances?

Reuters could not determine whether the bank balances that Meta Muse sent to employees belonged to other users or were invented by the model. Meta spokesperson Daniel Roberts called the cases isolated incidents and said many of the flagged account mix-ups were hallucinations that had already been addressed. Meta declined to say how many such issues there were or why they continued.

### Can Meta Muse read my Signal messages?

In tests by Reuters, the AI Now Institute and Signal staff, Meta Muse retrieved Signal message text on an Android phone through push-notification access, and on a Mac through accessibility and screenshot tools. The same requests were blocked on an iPhone. Reuters found no evidence that Signal's end-to-end encryption was broken; Muse read what the operating system already showed.

### How do I stop an AI agent like Muse from reading other apps?

Muse's Signal access on Android came from notification-reading permission, and on macOS from accessibility and screenshot permissions. Google says Android users must deliberately grant notification access, so revoking it in system settings removes that path. Apple added separate agent permissions in a Mac update this week, but permissions granted earlier stay in effect until a user removes them.

### What has Meta said it is doing about the Muse problems?

Meta says each Muse agent runs in its own secure virtual machine and that users choose which apps it can connect to. Its terms warn that Muse may make mistakes and make users responsible for the agent's actions. A member of Meta's Superintelligence team wrote internally that the company is working through problems in both the model and the software that lets Muse act.

## Links

- [TechCrunch Report](https://techcrunch.com/2026/11/19/metas-ai-agent-muse-is-mixing-up-conversations-and-exposing-private-data-report-says/)
- [Reuters Findings](https://www.ibtimes.co.uk/meta-muse-ai-signal-messages-bank-balances-1833200)
- [More Coverage](https://www.techjuice.pk/meta-muse-ai-privacy-signal-messages-mixed-users/)
