Abstract
This disclosure describes a method of deciding which sensor readings an AI governance and safety system may trust, and for what. Every certified sensor belongs to one of three trust classes. Class I readings are informational and may be used for operation and analysis. Class II readings are signed at the source with the sensor's hardware-bound identity and a trusted timestamp, and each is linked to the signed calibration record that governed it, so that the reading can serve as provenance and evidence. Class III readings meet Class II and are also independent of the AI system they watch: the sensor has its own power, its own communication channel and its own firmware, and the governed AI system cannot configure, silence or alter it. Only Class III readings may trigger an independent safety reset of the AI system, and a reset requires agreement among several independent Class III sensors, so that one faulty sensor can neither cause nor prevent it. A Class III sensor sends a continuous health signal, and loss of that signal is treated as an alarm, never as "all clear". Readings of one event from several sensors may be combined by an identified fusion node into a signed bundle that also records which expected readings were missing or unhealthy. A reading whose calibration record is missing, expired or unsigned is not accepted as evidence or as a safety input.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Wang, Edward D. H., "SI Sensor™: Sensor Trust Classes in which Only Sensors Independent of the Governed AI System May Trigger Its Reset, with Calibration-Linked Signed Readings and Fusion Bundles", Technical Disclosure Commons, (October 13, 2026)
https://www.tdcommons.org/dpubs_series/12190