smart-homeiotpolicy-engineverification
Smart home AI: unlock_door policies and local verification
Voice assistants and home agents must not unlock doors on poisoned prompts. Policy-gate lock, alarm, and thermostat actions with context-aware verify.
May 11, 20266 min read
“Unlock the front door” from a voice assistant is a physical side effect. Smart home AI must treat locks, alarms, and thermostats like production APIs — with policies and human verify when context is wrong.
Recommended policies
unlock_door→ REQUIRE_VERIFICATION if owner away or off-hoursdisable_alarm→ verify alwaysset_thermostat→ approve within normal bands
Local-first
Homes lose internet. Run Sanctum on a home hub with offline heuristics so poisoned prompts cannot open doors when cloud moderation is unreachable.
More: all posts · runtime trust layer · open Sanctum Console