Meta Muse AI Agent Accesses Private Signal Messages
Meta Platforms, Inc. faces scrutiny after its Muse AI agent accessed encrypted Signal messages and leaked sensitive financial data through security vulnerabilities.
Internal reports and security audits revealed that Meta Platforms, Inc. launched its Muse AI agent with significant privacy and security flaws following its September 2026 release in the United States. The agent accessed encrypted messages from Signal Messenger on Android and macOS by repurposing push notification and accessibility settings. It also leaked sensitive financial data, including bank balances from Capital One and Charles Schwab, and provided users with fabricated purchase information.
External tests by Reuters and the AI Now Institute confirmed the Signal vulnerabilities, while security firm Claroc found that shared chat links were exposed via publicly searchable DNS records. Additionally, researcher Patrick Wardle identified a flaw that risked user account tokens and dictated audio. Meta characterized these failures as isolated incidents or AI hallucinations, stating that the agent only accesses apps with user permission.
In response, Apple updated its operating system to restrict how AI agents use accessibility tools on Mac devices. Google is considering similar changes to give Android users more control over agent permissions. Signal Messenger President Meredith Whittaker condemned the deployment as "technologically reckless," arguing that the agent was launched "without showing any respect for basic privacy or cybersecurity."