Zero-Day Vulnerability in Meta's Muse App for macOS Exposes Linked Applications

Security researcher Patrick Wardle discovered a zero-day vulnerability dubbed "Not-a-Mused" in Meta's Muse app for macOS, which allowed attackers to hijack the AI agent and access linked accounts like WhatsApp, email, and calendar. Meta quickly pushed a hotfix removing the insecure voice dictation configuration setting responsible for the exploit. The issue highlights a significant new security threat vector associated with personal AI agents that aggregate user permissions across multiple services. By compromising the central agent, attackers gain indirect access to connected applications without needing to breach those apps individually. The exploit relied on an undocumented configuration parameter (`endo_voyager_dictation_endpoint`) that could be modified by any local process without triggering macOS permission prompts, leaking authentication tokens to an attacker-controlled server. Attackers could initiate this exploit using ClickFix social engineering, tricking users into pasting and running terminal commands under the guise of fixing software issues.