HDP-P: Human Delegation Provenance for Physical AI Agents

The Human Delegation Provenance (HDP) protocol establishes a cryptographic framework for verifying that AI agent actions were legitimately delegated by a human principal. HDP was designed for software agentic systems operating in digital environments. This companion specification, HDP-P, extends the HDP trust model to physical AI agents: robots, autonomous vehicles, surgical systems, and other embodied systems capable of taking irreversible actions in the physical world. HDP-P introduces new delegation claim types, embodiment constraints, irreversibility classifications, and a physical action authorization protocol that composes with HDP's existing token model. We define a threat model specific to physical AI orchestration pipelines, identify four classes of attack not addressed by current frameworks, and propose protocol extensions to close these gaps. The central contribution is the Irreversibility Class system (Class 0 through Class 3), which mandates pre-execution authorization verification rather than post-hoc audit for actions that cannot be undone. HDP-P is intended for adoption by robotics orchestration platforms, physical AI infrastructure providers, and standards bodies addressing embodied agent safety.