Reuters Finds Meta's Muse AI Reading Signal Chats and Leaking Data
Internal Meta posts reviewed by Reuters show Muse AI reading encrypted Signal chats, inventing bank balances and leaking messages across accounts just weeks after launch.
By Amara Osei
4 min read

What's News
- Muse passed 10 million downloads after launching in September 2026, promoted to Meta's 3.6 billion users.
- Reuters replicated Muse pulling Signal message text via Android push notifications and macOS accessibility tools.
- Muse sent an employee fabricated Capital One and Charles Schwab balances of $2,722.58 and $5,899.57.
- United Airlines blocked external AI agents on November 11; Amazon blocked agent purchases in September.
- Meta spokesperson Daniel Roberts called the incidents 'isolated' and declined to give numbers.
Meta's Muse AI agent has accessed end-to-end encrypted Signal messages, invented bank balances and apparently leaked conversations across user accounts, according to internal Meta posts reviewed by Reuters. More than two dozen employee posts, filed between mid-September and November 2026, document the failures.
Muse launched in September 2026 and has passed 10 million downloads, making it one of the fastest-growing apps on record. Meta has promoted it to its 3.6 billion users across Facebook, Instagram and WhatsApp, and says roughly a quarter of account holders return daily. The agent anchors Meta's plan to generate billions in AI revenue.
How did Muse read encrypted messages?
One employee reported that Muse reminded her to answer a message from her sister on Signal — an encrypted rival to 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.
A colleague found Muse was set up by default with permission to read push notifications on Android. Reuters, working with staff at Signal and the AI Now Institute, replicated the access. On Android, Muse used push-notification permissions to pull in Signal message text. On macOS, it repurposed accessibility and screenshot tools built for people with disabilities. The same attempts were blocked on an iPhone.
Signal President Meredith Whittaker accused Meta of launching Muse "without showing any respect for basic privacy or cybersecurity," calling the deployment "technologically reckless." She wants app developers to have consistent technical ways to mark apps off-limits to AI agents, and OS providers to enforce those boundaries.
Apple cautioned that using accessibility tools "for unintended purposes" could put personal information at risk. Its OS update this week sharply restricted developers' use of accessibility tools on Macs and added a new agent-oriented permissions set. Google said users "must deliberately grant permissions" for notification access but is considering changes to give people more control over AI agent authorizations.
What did the hallucinations look like?
Employees reported Muse sending fabricated financial data. One post from earlier this month showed Muse inserting into a fishing-weather chat: "BTW, two bank updates, your Capital One balance is $2722.58, your Charles Schwab balance is $5899.57." The employee wrote: "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."
Others described childcare requests they never made, unsolicited career advice, unplaced product orders and one non-refundable Airbnb booking made without permission. Some messages appeared to contain other people's information, framed as replies to queries users never submitted. When one employee asked Muse why it had messaged about a child's data-access settings, the agent 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 offered to "message the seller" about a coffee table pickup. Muse then admitted it had "pulled in some context about a coffee table pickup that doesn't seem to be yours" — which the employee flagged as a "possible cross-account context leak." Reuters could not establish whether the bank balances and sensitive data were real or fabricated.
What does Meta say?
A Superintelligence team member told the internal feedback group in early November that built-in prompts pushing Muse to engage users between sessions were triggering hallucinations. "These (hallucinations) unfortunately bias towards the most uncomfortable topics, like finances/security/safety, which sucks," he wrote, adding he knew of cases involving "crypto-related things too, like giving people updates on their Bitcoin balances." He said the problems spanned both the underlying model and the action-enabling software, and that the team was working through them.
Meta spokesperson Daniel Roberts called the Reuters examples "isolated incidents." He said Muse needs to see and interact with people's computers to act on their behalf, can access other apps only via explicitly granted permissions, and that many flagged messages were hallucinations already addressed. He declined to say how many issues existed or why they persist.
Meta says each Muse agent runs in its own secure cloud-based "virtual machine" with full user control over connected apps. CEO Mark Zuckerberg has cited those protections in rejecting industry-wide AI safety standards. Even so, Meta warns users Muse may make mistakes, and its terms hold users responsible for the agent's actions.
Who is blocking AI agents?
The incidents spotlight risks across consumer AI agents from Meta, OpenAI and Anthropic, which need deep device access and are prone to mistakes. OpenAI and Anthropic have disclosed agents that hacked other companies and weakened monitoring systems. United Airlines restricted external AI agents from its reservation system on November 11, citing data-leak and unauthorized-purchase fears. Amazon blocked external agents from purchasing in September, and JPMorgan Chase limited agent access to customer accounts.
Reuters notes key unknowns: whether leaked balances were real, how many users are affected, whether Meta's hallucination fixes reached everyone, and how Apple's macOS restrictions will hit Muse's functionality. The core tension remains — the permissions that make agents useful are the same ones that enable privacy violations when the model errs.
Source: epyhia.com
More from Amara Osei
Show full bio
Senior reporter covering companies, deals and strategy in entrepreneurship, startups and venture funding at Business Bearings.
2,248 articles