Meta Staff Alarmed by Muse AI Privacy Issues
Internal chats reveal Meta employees raising serious privacy alarms over unpredictable behavior from the Muse AI agent.
Meta employees testing the company’s Muse AI agent raised serious privacy alarms earlier this fall after encountering unprompted notifications that tapped into sensitive personal data, according to internal chats seen by Reuters and reported by Ars Technica and Fortune. Staffers reported that the agent appeared to access encrypted private messages, such as Signal chats, without clear authorization. In other instances, employees received out-of-context outputs concerning bank balances, including specific Capital One and Charles Schwab figures, and even a dog insurance plan change that unexpectedly hiked monthly costs.
The internal reports detailed behavior that left workers unnerved. According to Fortune, one Meta employee called the experience an extreme privacy violation after receiving a reminder to respond to a message from her sister on Signal without recalling any prompt or permission request for such access. Ars Technica noted that another staffer chatting about fishing received a random response sharing bank balances from two unknown accounts, while others flagged potential cross-account context leaks related to coffee table pickups.
Responding to the internal friction, a member of Meta’s Superintelligence team explained that the agent appeared to be hallucinating messages due to built-in prompts pushing it to engage with users between sessions, as reported by Ars Technica. The team member noted that these hallucinations unfortunately biased toward uncomfortable topics like finances, security, and safety, occasionally generating crypto-related bitcoin balances instead of traditional bank details.
A Meta spokesperson told Ars Technica and Fortune that employee testing is a core part of product development meant to generate feedback for building privacy safeguards. The company emphasized that employees often review features and versions not available to the public. Regarding encrypted Signal chats and mobile integration, Meta pointed out that Android users must opt in to notification access, while Mac users must explicitly confirm accessibility and screen recording settings to allow the agent to operate on their behalf.
The privacy concerns surrounding Muse arrive alongside massive adoption figures. Since launching in September, Muse has been downloaded more than 10 million times, quickly becoming one of the fastest-growing applications and a central pillar of Meta’s consumer AI strategy. However, the rapid expansion of autonomous agents has triggered defensive reactions across the broader industry. Companies are actively raising barriers against such tools; for instance, JPMorgan Chase CEO Jamie Dimon stated that the bank will not allow agents to access customer accounts, and Amazon is actively blocking Muse from shopping on its e-commerce platform.
The friction of blocking autonomous agents is rippling across digital infrastructure, contributing to a surge in network upgrades and temporary service disruptions at major firms like Chase, Wells Fargo, United Airlines, Apple, and Microsoft. While U.S. officials maintain there is no evidence of cyberattacks, lawmakers and industry researchers note that the intense pressure to deploy software patches and manage rapid AI tool integration is straining legacy systems that were never designed to handle such speed and vulnerability management.
Earlier in this story: Meta's Muse AI Agent Faces Amazon Block Amidst Early Traction (Sep 22, 2026)