WIRED: Meta's Muse Agent Excels at Data Collection More Than Getting Tasks Done
mmitchell_ai · x · 2026-09-22
A WIRED hands-on review of Meta's new personal AI agent app Muse, co-signed by researchers mmitchellai, evijit and samirpassi. Muse promises to handle deals, reservations and inbox management, hit 900,000+ downloads in its first week (Sensor Tower), and works via WhatsApp with a friend-like texting UX.
Key findings:
- The app aggressively pushes users to connect email, bank accounts, even passport info
- It continues Meta's pattern of opt-in data collection for AI training
- The author concludes it is more obsessed with collecting data than completing tasks
- Meta's spokesperson disputes this, citing built-in protections and user controls
The piece frames a core tension for consumer AI agents: mainstream versions of Silicon Valley-style agents (compared to OpenClaw and Instinct) may ship with immature task abilities but full-throttle data appetite.
More from Safety
- Margaret Mitchell: text watermarking arrived five years too late as AI slop floods the web — mmitchell_ai · 2026-09-22
- PIBBSS talk: "Stop if we catch AIs scheming" is no plan for automated alignment — JacquesThibs · 2026-09-22
- AI safety researcher: NVIDIA is 'reasonably close to maximally bad' on safety, but OpenAI is not — davidmanheim · 2026-09-22
- Fake Hugging Face org 'TypeSafeAI' posts 604B 'Step-5-Preview' model impersonation — realmrfakename · 2026-09-22
- Margaret Mitchell's new paper argues AI agents are pushing humans out of the loop — mmitchell_ai · 2026-09-22
- AI safety researcher David Krueger calls for immediate international moratorium on frontier AI — DavidSKrueger · 2026-09-22