Abstract
This disclosure describes a sensor for ambient AI systems, such as AI devices in homes, offices and shared spaces that watch and listen continuously, which enforces privacy and visibility rules in its own hardware and firmware, so that they hold even if the software above it is careless or compromised. Each sensor carries a signed declaration of what it senses, at what resolution, for which purposes and for how long anything is kept; its firmware cannot sense beyond the declaration. Processing happens on the sensor: it emits declared events, such as "door opened" or "person fell", rather than raw audio or video, and holds raw data only in a short encrypted buffer whose key is destroyed when the buffer expires. The sensor's visible indicator is powered from the same supply as its sensing element, so the element cannot be powered without the indicator lit, and its physical off switch breaks that supply. The sensor detects when its indicator is covered or damaged and stops sensing until it is restored. It filters out people who have signalled that they do not wish to be processed before anything leaves the sensor, honours requests to forget recent capture by destroying buffer keys, signs every output with its identity and trust class, and reports tampering, covering, movement or failure rather than falling silent.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Wang, Edward D. H., "SI Sensor™: A Governed Sensor for Ambient AI that Declares What It Senses, Emits Events Instead of Raw Feeds, and Powers Its Indicator and Off Switch from Its Sensing Element", Technical Disclosure Commons, (October 13, 2026)
https://www.tdcommons.org/dpubs_series/12191