Meta Rolls Out Hotfix for Zero-Day Flaw in Muse AI Assistant for Mac

Meta has issued a hotfix for a zero-day vulnerability in the macOS version of its AI assistant Muse. The flaw, related to the app's dictation feature, could allow local malicious programs to change the server endpoint used to process audio, potentially hijacking users' audio and account tokens.

David Singleton of Meta Superintelligence Labs announced the fix in the early hours of Tuesday, saying the company had removed the setting that allowed the dictation endpoint to be changed. He noted that the attack required malicious code already running under the user's account, calling it a local privilege escalation rather than a remote exploit, with low practical risk.

The vulnerability was discovered by Mac security researcher Patrick Wardle, who disclosed it Monday. Wardle said he built proof-of-concept attacks that could write files to disk or take photos without alerting the user, and argued the flaw could be abused through ClickFix-style social engineering.

Muse launched on Sept. 8 as a personal AI agent that can handle tasks such as sending emails, booking travel, shopping, and tracking goals. Users grant it access to various apps and accounts, raising concerns about any vulnerability that could hijack the agent.