Meta's Muse AI agent faces zero-day exploit, raises security alarms

ICD: International Cyber Digest is an independent cybersecurity reporting account focused on AI and digital policy developments. It published the details of the Muse zero-day vulnerability and the former Meta security manager's warnings. Meta: Meta Platforms develops social platforms and AI technologies including personal agents. It launched its Muse AI agent in September 2026 with built-in isolation and approval mechanisms while facing reports of security issues from departing staff. Muse: Muse is Meta's personal AI agent that performs tasks across connected apps and services on users' behalf. Launched in early September 2026, it is the subject of a reported zero-day exploit that could enable unauthorized prompt access and data exposure on Mac systems. Security Design: Meta built Muse with dedicated isolated virtual machines and a separate Sentinel system to review and approve actions before they reach external services. Employee Feedback: Internal testing of Muse around launch time surfaced reports of guardrail bypasses and reliability problems according to employee accounts reviewed in recent coverage. Bug Bounty Program: Meta opened a public bug bounty for Muse vulnerabilities shortly after its launch to support responsible disclosure of issues like prompt injection.