SafeGate / Agent-Readable Trust Preview

AI will automate payments. SafeGate will automate trust.

SafeGate now exposes an agent-readable commerce verification capability where apps and AI agents can understand verified post-payment states, receipt proofs, evidence status, access state, and public verification boundaries.

This page began as a static public-safe design preview. The current live capability is the read-only SafeGate Commerce Outcome API; it is not an autonomous agent permission system, and it does not control wallets, move funds, or authorize spending.

SafeGate verifies what happened after payment. Payment is the trigger. Trust is the product.

Live Agent-Readable Verification Live Read-Only API No MCP Endpoint No Autonomous Execution Public Verifier Live Not Formally Audited
Main Review Hub V13 Controlled Hardening Backend Policy Simulation Fee Architecture Public Repository

Why Agent-Readable Trust Matters

Human reviewers read pages. Apps and AI agents need structured trust states. SafeGate's current role is to make verified post-payment outcomes understandable to software without exposing custody, secrets, private keys, or unsafe execution paths.

SafeGate will not let agents spend, unlock, finalize, or verify trust unless the underlying payment and trust state is safe.

Agent-Readable Trust Object Preview

{ "safeGateAgentReadableTrustPreview": { "version": "0.1-preview", "status": "SCHEMA_PREVIEW_WITH_LIVE_READ_ONLY_API", "agentExecutionEnabled": false, "mcpEndpointEnabled": false, "toolEndpointEnabled": false, "productionReadyClaim": false, "formalAuditClaim": false }, "trustState": { "invoiceState": "CREATED_OR_VERIFIED_OR_FINALIZED", "paymentState": "LOCKED_UNTIL_BACKEND_VERIFIED", "feeState": "NOT_REQUIRED_OR_REQUIRED_PENDING_OR_VERIFIED", "receiptState": "NOT_CREATED_UNTIL_VERIFIED", "evidenceState": "NOT_CREATED_UNTIL_VERIFIED", "accessState": "LOCKED_UNTIL_FINAL_STATE", "publicVerifyState": "SAFE_FALSE_UNTIL_RECEIPT_EVIDENCE_MATCH" }, "agentReadableDecision": { "canTrustPayment": false, "canTrustReceipt": false, "canTrustEvidence": false, "canUnlockAccess": false, "canShowVerifiedOutcome": false, "reason": "Preview only. Real backend state required." }, "boundaries": { "paymentProcessing": false, "custody": false, "escrow": false, "privateKeyCollection": false, "passphraseCollection": false, "autonomousSpendingPermission": false, "officialPiPartnershipClaim": false } }

Agent-Readable State Model

  • Invoice state
  • Backend-verified payment state
  • Fee-required / fee-verified state
  • Receipt proof state
  • Evidence record state
  • Access locked / unlocked state
  • Public verify result
  • Trust confidence label
  • Known limitation flags
  • Safe next action suggestions

What Agents Must Not Do Yet

  • No autonomous payment execution.
  • No autonomous access unlock.
  • No receipt creation without backend-verified state.
  • No evidence creation without backend-verified state.
  • No public verified outcome from unknown pairs.
  • No private key or passphrase request.
  • No custody or escrow behavior.
  • No MCP/tool endpoint exposure before hardening.

Agent-Readable Roadmap

Phase SafeGate Focus Agent Scope
Foundation Payment verification, receipt, evidence, access state, public verify No agent execution
Integrity Hash proof, tamper-evident receipt, public-safe evidence Agents may read proof format later
Pilot Merchant pilot pack, tester evidence, controlled review Agents may read public review status
Live Agent-Readable Verification Live read-only commerce outcome verification and boundary language Read-only API live
Controlled MCP / Tool Layer Dedicated safe backend, auth, rate limits, allowlisted operations Controlled read-only tools first

Safe Agent Principles

  • Read before act.
  • Verify before trust.
  • Fail closed on ambiguity.
  • Never unlock before final state.
  • Never expose secrets.
  • Never claim production without evidence.
  • Never treat simulation as real settlement.

Why This Differentiates SafeGate

  • Most payment apps show payment status to humans.
  • SafeGate can make post-payment trust readable to software.
  • AI agents will need safe trust signals before acting.
  • SafeGate provides a trust-state layer between payment, evidence, and agent-readable outcomes.

Current Boundary

This page documents the current read-only agent-verification boundary. SafeGate is agent-readable without exposing autonomous execution, unsafe endpoints, private data, or payment authority.

The current priority is to keep the commerce trust spine narrow and safe while expanding integration around the live Commerce Outcome API.