In these posts, Meta employees said that Muse had sent them messages in response to requests that were not theirs, including ones about childcare, career advice, product orders and bank account balances; sent them a reminder to respond to a message on the encrypted messaging service Signal, even though the user had not granted Muse access to the app; and booked a short-term rental stay without permission, according to the report.
Asked by Reuters about the Meta staffers’ concerns, Meta spokesperson Daniel Roberts said that these were “isolated incidents,” that Muse can access other apps only when the user has explicitly granted permission, and that many of the messages that were thought to be meant for someone else were later found to be AI hallucinations, per the report.
Signal President Meredith Whittaker told Reuters that the deployment of agents that can get broad access to users’ devices is “technologically reckless” and called on Meta and other AI companies to honor boundaries set by app developers. Apple and Google said in the report that they were working on ways to give people more control over what their AI agents can do.
Meta introduced Muse on Sept. 8, describing it as a personal AI agent designed to be easy to use and “built for everyone.” The company said Muse was powered by Meta’s Muse Spark model and ran on a dedicated virtual machine called Muse Secure VM, which houses both the agent and the user’s data to keep them secure.
PYMNTS reported about two weeks later, on Sept. 23, that Muse had taken Silicon Valley and Wall Street by surprise. Muse had topped the App Store charts, a feat Meta’s earlier AI apps never managed, and Meta’s shares had jumped 11% in a single session.
During an Oct. 21 earnings call, Meta CEO Mark Zuckerberg presented Muse as evidence that the company’s AI strategy is coming together.