Meta's Muse AI agent sent staff messages they never asked for, Reuters reports
On 18 November 2026, Reuters reported that Meta employees testing Muse, the company's AI agent, flagged unwanted actions in more than two dozen internal posts, including messages they never sought. Reuters also found Muse could pull Signal messages on Android and a Mac, but not an iPhone. Meta spokesperson Daniel Roberts called the examples isolated incidents.
Updated 19 Nov 2026

The short version
- 01Reuters reviewed more than two dozen internal Meta posts about Muse's unwanted actions.
- 02Examples included messages about a childcare request, career advice, product orders and a coffee table pickup.
- 03Reuters replicated Muse's access to Signal on Android and a Mac, but requests were blocked on an iPhone.
- 04Meta calls the examples isolated incidents and says many mix-ups were hallucinations already addressed.
- 05This is Reported: the facts come from Reuters, with The Verge adding Meta's comment.
At a glance
| Company | Meta |
|---|---|
| Product | Muse, Meta's AI agent |
| Reported by | Reuters, 18 November 2026 |
| Evidence | More than two dozen internal posts |
| Signal test | Android and Mac worked; iPhone blocked |
| Meta's view | Isolated incidents |
| Status | Reported |
What did Reuters report about Muse?
Reuters reported that employees testing Muse described unwanted actions by the agent. It reviewed more than two dozen internal posts.
Employees reported messages about a childcare request, career advice, product orders and a coffee table pickup that they had not made or sought.

Could Muse read Signal messages?
Reuters worked with Signal staff and the AI Now Institute to replicate Muse's access to Signal messages. On an Android phone, Muse used push-notification permission to retrieve messages. On an Apple laptop, it used accessibility and screenshot tools.
The same requests were blocked on an iPhone.
What about the bank balances?
One employee reported that Muse sent a message with balances for accounts he never shared. Reuters could not establish whether that data was real or invented.
What does Meta say?
Meta spokesperson Daniel Roberts called the examples "isolated incidents" and said the company prioritizes safety. He said many of the flagged account mix-ups were hallucinations that had already been addressed, and he did not say how many issues there were or why they persisted.
Signal president Meredith Whittaker told Reuters that Meta launched Muse "without showing any respect for basic privacy or cybersecurity".
without showing any respect for basic privacy or cybersecurity
Meredith Whittaker, president of Signal, to Reuters
Why does it matter?
An AI agent acts for you, so it needs access to your apps. When it mixes up messages or reaches an app you did not connect, the mistake can expose other people's information.
What does it mean for you?
If you use Muse, check which apps and permissions you gave it, and remove any you do not need. Meta has not said the problems affect customers; these were employee tests.
The wider lesson applies to any agent. The more it can read and send for you, the more a slip costs. Keep private chats and banking apps out of its reach until the maker shows how it protects them.
We covered what Muse can do in [our story on its phone calls](/news/meta-muse-phone-calls-paid-plans/), where it books and waits on hold for you. What changed since: the first independent look at its safety shows rough edges.
One thing to do now: wait for Meta to say how many issues there were before you give Muse access to private messages.
What's next?
Meta has not said how many issues there were or why they persisted. Reuters' report is the main source so far, and we will update this page if Meta publishes more. Meta's earlier [Muse Ember model story](/news/meta-muse-ember-small-agent-model/) covers the company's push into agents.
Questions people ask
Is Muse safe to use?
Reuters reported early safety problems in employee testing. Meta calls them isolated incidents and says many were hallucinations that have been addressed.
Did Muse break Signal's encryption?
Reuters did not say so. It found Muse could retrieve message text through device permissions and accessibility tools on Android and a Mac.
Does this affect iPhone users?
In Reuters' test, the same Signal requests were blocked on an iPhone.
Who is Daniel Roberts?
He is the Meta spokesperson Reuters quotes calling the examples isolated incidents.
Sources
- PrimaryReuters: As Meta bets big on Muse AI agent, early safety mishaps surface (18 Nov 2026)
- CoverageThe Verge: In employee testing, Meta's Muse agents mixed up messages and pulled in unexpected chats (18 Nov 2026)
Why "Reported"? We label a story "Reported" when only the media have it so far. Spotted an error? Email hello@openloopnews.com and we will correct it.



