
Meta’s Muse AI Exposes Internal Files and Security Flaws in Rapid Succession
Meta’s new AI assistant, Muse, has faced two major security disclosures this week. First, researchers found a zero-day vulnerability allowing attackers to hijack the agent via a transcription endpoint flaw. Second, developers demonstrated that Muse’s cloud-based virtual machines can be easily coerced into leaking their entire filesystem, including internal code and documentation. Meta dismissed the filesystem leak as 'intended behavior' but issued a hotfix for the zero-day.











