Meta's Muse Agent: Wide-Open Backdoor on Mac

🔥 What happened Security researcher Patrick Wardle hijacked Meta's macOS AI agent Muse via a ClickFix attack, exploiting an undocumented function called endo_voyager_dictation_endpoint. Meta has since removed the feature but denies it was a remote exploit. 💡 Why it matters The app touches emails, calendars, mic, and camera — Wardle found over 50 commands attackers could use to leak audio, trigger prompt injections, or steal auth tokens. Amazon blocked Muse from its store shortly after launch. ⚡ Our take Giving an AI agent full system access and shipping an undocumented debug endpoint isn't a security bug — it's a security philosophy. Meta should kill Muse, not patch it.