Cryptographically Verifiable Inference Chain for AI Agent Computational Provenance

This document defines the inference_root claim as a companion to the actor_chain claim ({{!I-D.draft-mw-spice-actor-chain}}) and the intent_root claim ({{!I-D.draft-mw-spice-intent-chain}}). While the actor chain addresses delegation provenance (WHO) and the intent chain addresses content provenance (WHAT), the inference chain addresses computational provenance (HOW) — providing cryptographic proof that a claimed AI model actually performed the inference that produced a given output. The inference chain leverages two complementary mechanisms: Zero- Knowledge Machine Learning (ZKML) proofs for mathematical certainty, and Trusted Execution Environment (TEE) attestation quotes for production-scale AI workloads. The full inference chain is stored as ordered logs, with only the Merkle root included in the OAuth token for efficiency. Together, the three chains — actor, intent, and inference — form a complete "Truth Stack" for autonomous AI agent governance.