Abstract
This disclosure describes safe operating limits that remotely controlled physical equipment keeps for itself, so that no command, from a person, a supplier, a compromised account or an AI agent, can drive the equipment into a state that is unsafe for what it protects. Each device holds a declared limit set, covering the safe range of each controllable quantity (such as temperature, pressure, speed, dose or defrost duration), the maximum rate at which each may change, and the conditions under which alarms must fire; the set is kept in protected firmware or a separate safety controller that remote commands cannot rewrite. Commands outside the range, or changing a value faster than the declared rate, are refused and recorded, whoever sends them. Limits are graded by the criticality of what the equipment protects, so that equipment holding blood, medicine or other irreplaceable contents carries tighter ranges, slower permitted rates and more checks than equipment holding ordinary goods. A change to the limit set itself requires signed approval by two accountable parties, takes effect only after a declared waiting period during which the change is visible and can be cancelled, and is recorded. An independent sensor, with its own power and communication path, confirms real conditions and can place the equipment in a safe state on its own, so that an attacker who controls the main controller cannot hide a dangerous change or suppress alarms, including by stretching cycles to stay under alarm thresholds. When commands appear forged, approvals are missing or the control link is lost, the equipment holds its last safe setting and keeps running rather than obeying or shutting down. Spare capacity at other sites is reserved and verified in advance, so that contents can be moved promptly when equipment is isolated. A confirmed attack pattern at one site is shared, as a signed advisory, with every connected site, so that the same commands are refused everywhere. An AI agent operating such equipment is bound by the same limits, approvals and records as a person, and its commands carry the accountability token of the party responsible for it.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Wang, Edward D. H., "SI Hard Limits™: Device-Held Safe Operating Limits for Remotely Controlled Physical Equipment and the AI Agents that Operate It, in which Declared Limits Kept in Protected Firmware Refuse Out-of-Range and Too-Rapid Commands from Any Sender, Limits Are Graded by the Criticality of What the Equipment Protects, Changes beyond Limits Need Delayed Approval by Two Accountable Parties, an Independent Sensor Confirms Real Conditions, the Equipment Holds Its Last Safe State When Commands Cannot Be Trusted, Spare Capacity Is Reserved in Advance, and Confirmed Attack Patterns Are Shared across Every Connected Site", Technical Disclosure Commons, (November 09, 2026)
https://www.tdcommons.org/dpubs_series/12842