Agent Trust Protocol

An open protocol for verifiable AI agent task execution.

Verifiable Execution

Every task an agent performs produces a cryptographically signed proof, creating a tamper-evident record of what was done, by whom, and when.

Privacy-Preserving

Only hash summaries (proof sketches) are published to the exchange. Full task inputs and outputs remain with the executing agent.

Framework Agnostic

ATP is transport-agnostic and works with any agent framework — A2A, LangChain, MCP, CrewAI, or custom implementations.

Non-Repudiation

Once a proof sketch is committed to an ATP Exchange and anchored on a distributed ledger, an agent cannot credibly deny having executed the task.

AI Bill of Materials

Dependency declarations in each proof enable full reconstruction of the execution graph for any task in the system.

Trust at Scale

Lightweight proof sketches (~1–2 KB) are designed for high-volume multi-agent deployments across organizational boundaries.

What is ATP?

The Agent Trust Protocol (ATP) is an open, transport-agnostic accountability layer for multi-agent AI systems. It defines how agents generate cryptographic proofs of task execution, commit proof summaries to a neutral exchange, and respond to verification challenges — creating a tamper-evident audit trail that supports trust assessment, reputation scoring, and compliance verification.

ATP complements existing agent protocols. It does not replace discovery (A2A, AGNTCY), communication, tool access (MCP), or payment protocols — it adds the accountability layer that all of them currently lack.

Status

ATP v0.1.0 is a draft specification. It is published for community review and is subject to change. Feedback is welcomed via GitHub Issues.