Agents are moving from answering questions to executing real operations:
Action systems need verifiable receipts, not only conversational transcripts.
Traditional logs are useful for debugging, but weak for portable trust:
For high-risk operations, logs alone are insufficient for independent verification.
TrustProof uses signed action receipts with deterministic verification:
jti)prev_hash / entry_hash)Any verifier with the public key can independently check signature and integrity.
allow vs step_up decisions were produced.TrustProof keeps verification portable across execution platforms and organizations.
The agentic commerce ecosystem reached an inflection point in Q1 2026:
Mastercard Verifiable Intent (March 5, 2026): Mastercard launched cryptographic purchase receipts — proof that a human authorized a specific transaction. This solves payment fraud for agent commerce but leaves a gap: who authorized the agent to act in the first place?
Google AP2 Agent Payments Protocol (60+ partners): Google’s Agent-to-Pay Protocol defines how agents request, approve, and execute payments with 60+ launch partners. AP2 handles payment authorization; it does not handle the trust decision that precedes payment.
Coinbase Agentic Wallets (50M+ transactions): Coinbase’s agent-native wallet infrastructure processes tens of millions of on-chain transactions. The x402 protocol handles crypto payments, but agent authorization remains application-specific.
Nvidia NemoClaw / OpenClaw: Nvidia’s open-source agent sandboxing framework (NemoClaw, now OpenClaw) provides runtime guardrails — file access, network calls, tool invocation. OpenClaw handles what an agent can do. TrustProof proves what an agent was allowed to do and records the outcome.
Karpathy’s autoresearch (42K stars): Andrej Karpathy’s autonomous research agent demonstrated that complex, multi-step agent workflows need more than logging — they need cryptographic audit trails that survive across sessions and organizations.
AP2 handles payment authorization. Verifiable Intent handles purchase proof. NemoClaw handles agent sandboxing. TrustProof handles everything else — the 99% of agent actions that are not payments.
TrustProof v0.2 is the universal trust layer that connects these protocols:
protocol_refs.verifiable_intent_id → links to Mastercard’s payment proofprotocol_refs.ap2_mandate_id → links to Google’s payment mandateprotocol_refs.x402_payment_hash → links to Coinbase’s on-chain proofprotocol_refs.mcp_tool_call_id → links to Anthropic’s tool invocationsprotocol_refs.upstream_proof → links to other TrustProofs across platformsThe result: a single, verifiable receipt that proves an agent was authorized to act, what it did, and how that action connects to payment authorizations, tool calls, and other protocol artifacts across the entire agent commerce stack.