Cybersecurity researcher Patrick Wardle has uncovered a critical zero-day vulnerability in Meta’s macOS AI assistant Muse that permits locally installed applications or terminal commands to gain complete control over a user’s account. Introduced recently, Muse handles tasks such as booking appointments, making purchases, and connecting to services by accessing system permissions, email, calendar, and social media accounts.
The security flaw allows any process on the device, regardless of its macOS permissions, to alter Muse’s settings and hijack authentication tokens. Attackers can exploit this by redirecting the endpoint where speech transcription occurs to an attacker-controlled server, granting full access to user tokens and enabling malicious file generation or unauthorized photo capture without alerting the user.
Meta has not yet responded to queries regarding the flaw, which comes amid growing concerns over AI assistant security and access privileges. The discovery challenges Meta’s claims that Muse was built from the ground up for privacy and security, as Amazon has simultaneously moved to block the assistant from its site.
Why it matters
Highlights serious security risks associated with OS-level permissions granted to privileged AI assistants.
Demonstrates how credential hijacking in AI tools can bypass traditional operating system security boundaries.
Source: arstechnica.com



