Meta Patches Muse Vulnerability That Allowed Hijacking of AI Agent

Meta released an emergency hotfix for a zero-day flaw in its Muse macOS application after security researcher Patrick Wardle discovered that attackers could take complete control of the AI agent. The flaw stemmed from an undocumented setting that enabled local code to redirect cloud transcription processing from Meta's servers to an attacker-controlled endpoint, granting access to the user's Muse account. Wardle's proof-of-concept exploit demonstrated that the flaw allowed malicious programs to leverage Muse's privileges to write files to disk and take photos without alerting the user. While David Singleton of Meta Superintelligence Labs noted that the risk was minimal due to requiring local access under an existing user account, the incident has highlighted security oversights in the design of the AI assistant shortly after its launch.