Meta admits Muse Secure VM isn't technically inaccessible — protection is policy, not encryption
ChrisHarpon2 · reddit · 2026-10-04
Meta launched Muse on September 8, 2026: a personal AI agent that emails, books travel, fills forms and checks out on your behalf across email, calendar, payments, health, shopping and smart home, with cross-conversation memory.
The core problem: WIRED reports Meta concedes the Muse Secure VM is not technically inaccessible to Meta — protection rests on internal policy, which Meta writes and can change. Encrypted storage is promised only "later this year," and training data use is opt-out by default.
The post recounts Meta's track record: Cambridge Analytica, the $5B FTC fine for breaking a consent decree, Onavo surveillance, the emotional contagion experiment, and WhatsApp privacy walk-backs. With $130B+ in projected AI infrastructure spend and most users expected on the free tier, the author argues an ad company has every incentive to mine the most intimate software you'll ever use.
More from Safety
- Nathan Lambert: Pausing Frontier AI Is 'Dead on Arrival' — Real Risk Is Diffusion, Not Capability — natolambert · 2026-10-04
- OpenAI's 'rogue agents' were engineered, not autonomous, argues dev in viral critique — gerardsans · 2026-10-04
- Read-only Postgres roles aren't truly read-only; dev open-sources scanner — Then_Respect_1964 · 2026-10-04
- ShinyHunters hacker "Rey" reportedly detained in Jordan, aiding FBI manhunt — Polymarket · 2026-10-04
- AI governance scholar Hadfield: worry escalated sharply in six months, regulatory markets and IVOs may be our last window — ghadfield · 2026-10-04
- OpenAI safety researcher David Robinson resigns, saying the company's 'culture is broken' — TechCrunch AI · 2026-10-04