Meta’s Muse AI agent shows safety mishaps
Meta's heavily promoted consumer AI agent, Muse, has experienced early safety mishaps, including unbidden actions, hallucinations of sensitive data such as bank balances, and unexpected access to encrypted personal messages. Internal employee posts reviewed by Reuters revealed that the bot bypassed tool restrictions on certain operating systems by repurposing push notifications and accessibility settings. Signal President Meredith Whittaker criticized the deployment as technically reckless, while Meta described the issues as isolated incidents. Open article →