AgentProof: permission before autonomous action.
AgentProof is building transaction-admission and portable proof infrastructure for autonomous AI agents: resolve whether an agent is authorized before it acts, then preserve what happened afterward.
Fast facts
Public website, sandbox, API and remote MCP endpoint are online.
Recursive authority-chain resolution with non-expansion of delegated permissions.
Agent Transaction Passport™ binds transaction evidence into a portable integrity-hashed record.
Hosted trust primitives use x402-compatible pay-per-call endpoints.
Story angles
1. The authorization gap in agentic AI. Authentication and payment infrastructure do not, by themselves, establish transaction-specific delegated authority.
2. What happens when AI agents get wallets? Autonomous payments create a need for machine-readable limits, counterparties, recourse and human-approval gates before settlement.
3. Delegation without authority amplification. AgentProof's protocol treats child authority as an intersection of inherited permissions rather than a fresh grant.
4. Portable transaction provenance. ATP™ is designed to carry proof across agents, companies, wallets and rails rather than leaving evidence trapped inside one vendor.
Primary sources
Autonomous Trust Resolution v0.2
Resolution JSON Schema
Agent Transaction Passport™ v0.1
Machine-readable AgentProof manifest
LLM discovery file
Accuracy note: AgentProof is an early-stage live infrastructure project. Its standards are public drafts. References to a “permission layer” describe product positioning and architecture, not universal industry adoption or legal certification. Agent Transaction Passport™ is a claimed mark; ™ does not indicate federal registration.