Valid authorised action
An approved human and recognised agent propose an action inside the agreed amount, scope and time. Expected result: PASS, one-use GatePass, Trust Receipt and full audit receipt.
Agent Trust Gate™ is ready to support a suitable organisation in a bounded, non-production evaluation of one exact AI-agent action workflow. The buyer keeps execution control; ATG makes the pre-action trust decision and produces reviewable evidence.
An agent can be authenticated and still propose the wrong amount, supplier, account, scope or timing. The controlled pilot tests the trust boundary immediately before a consequential action without connecting ATG to a live execution system.
Synthetic or approved test action, authority, mandate, policy and agent-standing evidence.
Canonical action, authority, standing, limits, policy, freshness, signatures and replay state.
ATG issues a GatePass only for a fully allowed action, otherwise a recorded refusal.
The buyer owns every real execution decision and system. The default pilot performs no live action.
The pilot uses deterministic, bounded scenarios so business, security and control owners can inspect both successful and refused outcomes.
An approved human and recognised agent propose an action inside the agreed amount, scope and time. Expected result: PASS, one-use GatePass, Trust Receipt and full audit receipt.
Overspend, wrong authority, expiry or out-of-scope action. Expected result: REFUSE with an intelligible root cause, refusal evidence and no valid GatePass.
Change the action after approval, alter a bound value or reuse a consumed GatePass. Expected result: invalid execution authority with tamper or replay evidence.
Buyer-controlled test inputs → ATG PASS / REFUSE + evidence → buyer retains execution control.
The default pilot is decision-and-evidence only. No buyer execution system is connected. ATG does not create a real order, payment or external action.
The controlled pilot is deliberately narrow. It is intended to establish whether the control adds value before any separate production engineering or assurance programme is considered.
No production credentials, payment tokens, bank details, live customer data or live execution endpoints are permitted in the standard pilot.
The pilot stops for an authority bypass, inconsistent decision, exact-action binding failure, replay failure, unexplained receipt mismatch or need for live money movement.
The buyer remains authoritative for identity, permissions, policy, legal/compliance approval, test data, environment and any downstream execution.
P3-M156 independently assessed the practical pilot boundary rather than simply carrying forward the prototype hardening score.
Ready for a bounded controlled buyer pilot does not mean production ready.
ATG does not claim certification, regulatory approval, guaranteed compliance, guaranteed fraud prevention, live payment processing, production deployment or commercial validation.
Tell us the consequential AI-agent action you want to test. A synthetic or properly sanitised example is enough for the first discussion.