Science & Tech · AI · 7 hrs ago
Meta employees raise privacy concerns over AI agent Muse
Meta employees say its AI agent Muse sent messages they never asked for, including career advice, childcare replies and purchase details.
Some messages included bank balances employees said were not theirs, though it is unclear whether Muse exposed real account data or made the figures up.
Muse launched in September and has passed 10 million downloads, as Meta positions it as a tool that can carry out tasks for users.
In tests on Android phones and a Mac, researchers found Muse could use notification permissions to read message text, including Signal messages; the tests did not work on an iPhone.
Meta says the incidents were isolated or involved made-up information and says it is working on safety and privacy protections.
The company also says employees can use versions of Muse with features not available to the public.
The reports raise concerns about what access users grant AI agents when they connect them to devices and apps.
Meta says it has addressed some flagged messages, but the available information does not establish whether users’ private data was actually exposed.
Meta employees have raised concerns that its AI agent Muse may be mixing up conversations and exposing private information.
Staff reported receiving unsolicited messages about childcare, career advice, purchases and bank balances they said were not theirs.
Reuters reviewed more than two dozen internal posts and worked with Signal staff and the AI Now Institute to test Muse’s access to Signal messages.
Meta said the incidents were isolated or involved hallucinations, and that it works to improve safety and privacy protections.
Muse launched in September and had more than 10 million downloads, according to TechCrunch and the Straits Times.
- Who
- Meta employees and the company’s AI agent Muse are at the center of the reports.
- What
- Employees reported unsolicited messages and possible exposure of private information; Meta disputed that the examples showed a broader problem.
- When
- The reports were published on November 18 and 19, 2026. Muse launched in September.
- Where
- The incidents were reported in internal company posts and involved Muse on employees’ phones.
- Why
- Employees were concerned Muse could be mixing up conversations between accounts and exposing private data; Meta said some messages were hallucinations.
Meta
Concerned employees and Reuters investigation
Incidents
Meta
Meta described the examples as isolated incidents or hallucinations.
Concerned employees and Reuters investigation
Employees reported messages and actions they had not requested, raising concern that Muse could be mixing up accounts.
Access to other apps
Meta
Meta said Muse can access other apps only through permissions users explicitly grant.
Concerned employees and Reuters investigation
An employee said she did not grant Muse access to Signal; Reuters and partner testers reported replicating access to Signal messages.
Safety response
Meta
Meta said it conducts internal testing and continuously updates products to improve protections.
Concerned employees and Reuters investigation
Employees raised concerns through internal posts, and Reuters’ tests indicated the Signal access issue was ongoing.
Meta launched Muse.
An employee posted that Muse had added bank balances to a chat about fishing weather and tide information.
The Straits Times published Reuters’ findings on internal posts and testing of Muse’s access to Signal messages.
ZDNet and TechCrunch published reports on employees’ privacy concerns and Meta’s response.
- Downloads
- More than 10 million since its September launch
- Internal posts reviewed
- More than two dozen, according to Reuters
- Bank balances in one message
- Capital One: $2,722.58; Charles Schwab: $5,899.57
- Testing devices
- Android phone and Mac; the tests did not work on an iPhone, according to ZDNet
- Signal
- An end-to-end encrypted messaging service used in the reported incident
4 sources
Meta Muse booked an Airbnb without permission – and that’s not the worst of it
Meta’s AI agent Muse is mixing up conversations and could be exposing private data, report says
Muse privacy invasions are freaking out Meta staffers, internal chats show
As Meta bets big on Muse AI agent, early safety mishaps surface
What people are saying
isolated incidents
This seems like an extreme privacy violation. I definitely did not grant it access to other apps, or recall being asked about this
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
Employee testing is core to the product development process and the entire point is to get feedback, so we can implement safety and privacy protections and improve products