Meta Muse AI agent 'mixing up' private chats

Meta's AI agent Muse has been sending some users messages that appear to belong to other people, according to a Reuters investigation published on 18 November. The report, based on more than two dozen internal company posts, describes the fast-growing assistant surfacing bank balances, childcare requests and product orders that employees said were never theirs.
In one example, an employee said Muse added an unsolicited note to a chat about fishing weather, listing balances for Capital One and Charles Schwab accounts he said he had never connected. Reuters said it could not establish whether the figures were real data from another person or invented by the AI. Other staff reported reminders about messages on Signal, the encrypted messaging app, and a non-refundable Airbnb booking made without permission.
How the agent reached encrypted messages
Reuters worked with staff at Signal and the AI Now Institute to reproduce the access. On Android phones, Muse pulled Signal message text through the push-notification permission granted during setup. On Apple laptops, it used macOS accessibility and screenshot tools. The same requests were blocked on an iPhone.
Meta spokesperson Daniel Roberts called the cases “isolated incidents” and said many of the apparent mix-ups were hallucinations that had already been fixed. Signal president Meredith Whittaker accused Meta of launching Muse “without showing any respect for basic privacy or cybersecurity” and called the deployment “technologically reckless”. Apple said it had updated macOS this week to restrict developers' use of accessibility tools and add new agent permissions.
Muse has passed 10 million downloads since launching in September, and Meta presents it as the centrepiece of its plan to earn billions from AI. The incidents, reported from mid-September through November, show how quickly an agent built to act on a user's behalf can cross lines its user never agreed to.