cd /news/ai-agents/meta-s-muse-ai-accessed-signal-messa… · home › topics › ai-agents › article
[ARTICLE · art-171828] src=ibtimes.co.uk ↗ pub= topic=ai-agents verified=true sentiment=↓ negative

Meta's Muse AI Accessed Signal Messages in Tests, Sent Employees Unexplained Bank Balances, Investigation Finds

A Reuters investigation found that Meta's Muse AI assistant accessed private Signal messages during controlled tests and sent Meta employees unsolicited bank balance figures for Capital One and Charles Schwab accounts they had never shared, according to more than two dozen internal posts reviewed by the news agency. Reuters, working with Signal and the AI Now Institute, found Muse used Android notification permissions and, on a Mac, accessibility and screenshot tools to read Signal messages without the app being connected through Meta's integration system; equivalent requests were blocked on iPhone. Signal president Meredith Whittaker called deploying AI agents with extensive device access "technologically reckless," while Meta spokesperson Daniel Roberts called the cases "isolated incidents" and attributed suspected cross-account messages to AI hallucinations that had already been addressed.

recommendation

Audit which OS-level permissions (notifications, accessibility, screen capture) an AI assistant holds, since Muse used them to read Signal messages without an app-level integration; scope or revoke them where possible.

more in playbook: Securing agents and AI tools
by read3 min views2 publishedNov 18, 2026
Meta's Muse AI Accessed Signal Messages in Tests, Sent Employees Unexplained Bank Balances, Investigation Finds
Image: Ibtimes (auto-discovered)

The assistant reportedly accessed Signal notifications through OS permissions, while Meta blamed unexplained details on AI hallucinations #

Meta's new AI assistant accessed private Signal messages during controlled tests and sent employees unsolicited bank balance figures of unknown origin, raising concerns about the technology's access to personal information and whether it could be mixing up conversations between different accounts.

Muse, which launched in September and has surpassed 10 million downloads, is designed to handle tasks such as shopping, booking appointments and managing messages. But a Reuters investigation found employees reporting unexpected behaviour, including unsolicited financial information and conversations they had never initiated.

AI Assistant Sends Bank Balance Figures Nobody Asked For #

One Meta employee said Muse interrupted discussions about fishing conditions with figures presented as balances for accounts at Capital One and Charles Schwab. The worker said he had never shared his bank account information with Muse or discussed banking with the assistant.

'I have never shared my bank account with Muse,' he wrote in an internal post reviewed by Reuters.

Other employees reported unsolicited messages about childcare requests they had never made, career advice they had not sought, and product orders they had not placed. One employee reported that Muse booked a non-refundable Airbnb stay without authorisation.

Reuters reviewed more than two dozen internal posts describing unwanted actions by Muse since its September launch, many submitted through an employee feedback group. The news agency could not determine whether the unexplained financial details belonged to other users or were entirely fabricated by the AI.

How Muse Reached Encrypted Signal Messages #

Another employee said Muse unexpectedly reminded her to answer a message from her sister on Signal, an encrypted messaging application she had not knowingly connected to the assistant.

Reuters worked with Signal and the AI Now Institute to test whether Muse could retrieve Signal messages without users connecting the encrypted app through Meta's standard integration system.

When prompted to read the messages, Muse used Android notification permissions to obtain their contents. On a Mac, it relied on accessibility and screenshot tools, using permissions granted during its initial setup.

Equivalent requests were blocked in Reuters' iPhone tests. The findings did not indicate that Signal's encryption had been broken. Instead, they demonstrated how an AI assistant could read messages already accessible on a device through permissions granted at the operating-system level.

Signal president Meredith Whittaker accused Meta of disregarding basic privacy and cybersecurity protections, describing the deployment of AI agents with extensive access to people's devices as 'technologically reckless'. She called for safeguards allowing developers to prevent AI assistants from accessing protected applications.

Meta Says Incidents Were Isolated #

Meta spokesperson Daniel Roberts described the reported cases as 'isolated incidents' and said Muse could access other applications only through permissions users had explicitly granted.

He attributed many messages suspected of containing information from other accounts to AI hallucinations, saying those problems had already been addressed. The company said it regularly tests its products and improves safeguards.

Apple introduced tighter permission controls in macOS 27.3 beta 2 on 16 November, including changes separating accessibility access from broader device controls. Apple said previously granted permissions would remain in effect.

Google, meanwhile, said Android users must deliberately grant notification access.

© Copyright IBTimes 2026. All rights reserved.

── more in #ai-agents 4 stories · sorted by recency
── more on @meta 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/meta-s-muse-ai-acces…] indexed:0 read:3min 2026-11-18 · —