Abstract
This disclosure describes how AI systems and trained people share the work of keeping people safe in buildings, sites and public spaces, and how an alarm reaches the right people fast without exposing anyone more than needed. AI watchers, governed sensors and robots observe and raise alarms as signed events, but are never permitted to use force on, restrain, confine or confront a person; every action on people is taken by a human. A person in danger may raise an alarm from a personal device holding their portable credential, here called a passport, openly or silently under duress, and certified sensors and AI watchers may raise alarms on their own. Each alarm is a small signed record of its kind, severity, coarse location and expiry. The alarm trickles outward in rings: first to the person's chosen contacts and the on-duty responders of the site, then to the nearest certified human responders, then to site robots that may only observe, light the way or open exits within safe limits, then to emergency services through the site's operator, and finally, where people have opted in, to nearby willing helpers, who receive only a request for help and safe-route guidance. Each ring is reached only if the one before does not acknowledge within a declared time. At each hop, sender and receiver check each other's passports, live certification and mandate, so that alarms cannot be forged or injected, and repeated false alarms are traced to their source. Precise location and live footage are released only to responders who have accepted the alarm, under a signed, time-limited lease that ends when the alarm is closed. When networks fail, alarms continue by short-range relay from device to device and by store and forward, with a hop count and expiry, and are reconciled when connection returns. Every alarm, acknowledgement, handover, release of footage and closure is recorded in a signed incident record. Responders hold live certification as human responders, so that skill is proven continuously despite high staff turnover.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Wang, Edward D. H., "SI Alarm™: Governed Human–AI Safety Teaming, in which AI Watches and Alerts but Never Acts on People, Alarms Trickle Outward through Verified Passports to the Nearest Certified Human Responders First, Continue Offline Device to Device, and Every Handover Is Signed with Bystander Privacy Preserved", Technical Disclosure Commons, (November 02, 2026)
https://www.tdcommons.org/dpubs_series/12617