Deep-scan reveals Muse agent's hidden powers: Instagram DMs, HomeKit control, its own email identity
flaneur451 · reddit · 2026-09-23
- Reddit user flaneur451 ran a 90-minute 'deep self-scan' of the Muse agent, confirming via exposed tool schemas and benign probes a batch of capabilities absent from public docs — many deeply tied to the Meta hardware ecosystem.
Confirmed hidden capabilities:
- Full HomeKit control: rooms/accessories/traits, scenes, security sweeps, geofence-triggered automations; plus BLE scanning, Find Device for Ray-Ban Meta glasses, HealthKit historical backfills.
- Comms: real calls and SMS from a paired wearable (wearables-comms), Instagram and Threads DM sending (user-approved per message), and its own email identity (muse-mail); iPhone SMS is draft-only, never direct send.
- Others: custom voice creation (voice-design), bounded read-only SQL self-inspection of muse.db (private reasoning always redacted), a talking-head-to-vertical-video pipeline, and third-party integrations incl. Tesla, Philips Hue, Arlo, Etsy, Tailscale.
Notable: most of these are absent from all public sources (Patrick Wardle's Sept 21 security demo is the only external mention), showing consumer agents' actual permissions far exceed documented ones.
More from Models
- Theory: Typos Disrupt LLM Tokenization, Which May Explain Why AI Text Has a Distinct Smell — menhguin · 2026-11-11
- LMArena co-founder explains AI evaluation methodology and policy in new podcast — MattPerault · 2026-11-11
- Runway open-sources Praxis-1, turning video pretraining into real robot control — runwayml · 2026-11-11
- DeepMind launches AlphaGenome 2, spotting nearly 2x more gene-regulating mutations — GoogleDeepMind · 2026-11-11
- MazeBench: Fable 5.2 scores 15%, near-frontier at half price; Gemini 4 Carbon 11% — patience_cave · 2026-11-11
- ERNIE 5.5 Flash Preview enters LMArena Battle Mode for crowdsourced ranking — arena · 2026-11-11