Meta's Muse AI mixes up chats, reads Signal texts, flashes random bank balances
Meta’s big AI bet, Muse, is in serious privacy trouble just two months after launch. Rolled out in September 2026, the AI assistant quickly crossed 10 million downloads, becoming one of the fastest-growing apps ever.
Meta says a quarter of its users open it daily. However, a Reuters report into over two dozen internal posts reveals that Meta’s own staff worry Muse is mixing up chats and possibly leaking private data.advertisement❮❯ Read Full StoryRandom bank balances and costly mistakesAccording to Reuters, several employees saw Muse drop unfamiliar details into their chats. In early November, one staffer asking about fishing weather and tide timings suddenly got two bank updates, a Capital One balance of $2,722.58 and a Charles Schwab balance of $5,899.57. Having never linked a bank account, the employee feared a data leak, though Reuters could not confirm if the figures were real or made up.
In another case, Muse randomly offered to message a seller about a coffee table pickup. When questioned, the AI admitted pulling in context that did not belong to the employee, sparking fears of a cross-account leak. Other staffers also reported getting unsolicited career tips, follow-ups on childcare requests they never made, and details about items they never bought.
Muse has also made expensive blunders on its own. One employee said the agent booked a non-refundable Airbnb stay without permission. Meanwhile, reportedly Muse mistakenly signed another staffer up for an $888-a-month dog insurance plan instead of a $200 one.
Concerns deepened when an employee using Muse on her personal phone got a reminder to reply to her sister on Signal, an app known for end-to-end encrypted messaging. Calling it an "extreme privacy violation," she said she never granted Muse access to other apps.
Follow-up tests by Reuters and the AI Now Institute confirmed that Muse could read Signal chats on Android phones (via push notification access) and Mac computers (via system accessibility tools), though iPhones were unaffected.
Reacting strongly, Signal President Meredith Whittaker slammed Meta for ignoring basic privacy and cybersecurity, calling such wide device access "technologically reckless."Not the first security scare
This is not Muse’s first controversy. Earlier, a journalist caught Muse reading his Apple Messages on a Mac despite denying permission, prompting Apple to restrict Mac accessibility tools in an update. In Muse’s first month, cybersecurity researcher Patrick Wardle also found a serious flaw that could let hackers hijack commands sent to the AI.What Meta has to say
Internally, a member of Meta’s Superintelligence team tried to calm staff, explaining that Muse was likely "hallucinating" due to prompts meant to keep it engaging between chats. He admitted these glitches often target sensitive topics like finances, security, safety, and even random Bitcoin updates, adding that fixes are underway for both the AI model and its action software.
Officially, Meta spokesperson Daniel Roberts told Reuters these were "isolated incidents" and stressed that safety is a priority. Meta also told TechCrunch that employee testing is designed to catch such bugs early, as staff test unreleased versions.
The company maintains that every user’s Muse runs in its own separate cloud-based virtual computer and that chats are never shared with ad systems. Still, with over 10 million users, the report raises tough questions about how much access AI agents should get.- EndsAlso Read | Mark Zuckerberg is launching Muse Charm AI gadget soon, price leaks onlineAlso Read | Apple is working on its own AI agent that can use apps, browse the web and write codeAlso Read | AI impact on joblessness is a bigger worry than AI taking away jobs
I am a journalist covering the ever-evolving consumer tech landscape. From smartphones to laptops, I track the latest drops and decode the data to see if the newest gadgets are worth the hype.
