Abstract
UAIF‑2.0 defines a vendor‑neutral, machine‑readable safety contract and deterministic enforcement interface connecting AI agent governance systems to independently assured physical‑safety controllers. Existing AI governance layers authorize actions; existing functional‑safety systems enforce physical constraints. No broadly adopted, vendor‑neutral standard currently defines how authorized AI actions should be translated into bounded physical‑domain constraints consumable by a safety‑rated controller.
UAIF‑2.0 introduces a Safety Contract and a Safety Envelope Compiler. The contract binds AI identity, capability, authorization window, physical constraints, safety envelope, provenance, and revocation status into a deterministic, cryptographically verifiable document. The compiler transforms human‑authored safety rules and engineering constraints into a monotonic, traceable, fail‑closed machine‑readable envelope. The independent physical‑safety controller evaluates the contract against its own certified limits and real‑time environment state, ensuring that AI authorization can never expand the controller’s safety envelope.
UAIF‑2.0 defines the permitted runtime envelope as:
Permitted Runtime Envelope=Contract Envelope∩Controller Safety Envelope∩Current Physical-State Envelope
Supporting modules (UAIF‑1.4 Threat Containment, UAIF‑1.5 Global Safety Registry, UAIF‑1.6 Behavioral Fingerprinting) operate in the control plane to propagate revocation, detect anomalies, and maintain safety metadata.
UAIF‑2.0 is submitted as a proposed vendor‑neutral architecture and specification for defensive publication.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Caldwell, Michael Victor Mr., "UAIF‑2.0: Vendor‑Neutral AI‑to‑Physical‑Safety Contract and Deterministic Enforcement Interface for Cyber‑Physical Systems", Technical Disclosure Commons, (September 07, 2026)
https://www.tdcommons.org/dpubs_series/11621