Meta's Muse AI Agent Hits 3.4M Downloads and #1 on App Store—Then a VM-Exposure Bug Surfaces

Meta's aggressive consumer AI push is paying off and backfiring simultaneously. Its personal assistant Muse rocketed to 3.4 million downloads and the #1 spot on Apple's App Store following Meta Connect, where the company announced video chat with a Muse avatar, Mac computer-use capabilities, integration across 100+ AI-glasses styles, and a pocket-sized Muse Charm device shipping in December. Monetization spans $20/month and $100/month premium tiers.

The surge, however, collided with a sharp security disclosure. Researcher Patrick Wardle published a proof-of-concept showing malware could hijack Muse through an undocumented `endo_voyager_dictation_endpoint` setting, and a separate filesystem exposure let an attacker export 6.8GB of user VM data. Meta hot-fixed the issues after disclosure, but the timing—against a viral download surge and new computer-use features that operate on users' machines—amplified skepticism about whether Muse shipped before it was hardened.

The SEV-2 vulnerability matters more because of what Muse now does: with Mac computer-use, the agent operates a real desktop, so a hijack isn't a chatbot annoyance but potential system compromise. Nvidia even highlighted an engineering collaboration to make the Muse Realtime Avatar efficient enough to keep up with live conversation, underscoring how deeply integrated the product is.

Community reaction was pointed: Andrew Ng's widely-shared LinkedIn post referenced 'Meta's agent security' amid the broader AI-fear discourse, while security researchers argued the episode validates caution about consumer agents with local execution rights. The contrast is stark against the same week's OpenAI containment failures—two of the biggest consumer/enterprise agent launches both stumbling on security. Watch whether Meta's December Muse Charm hardware ships with tighter defaults.