Agent Trust Gate separates identity, standing, authority, approval, exact-action proof and execution evidence instead of collapsing them into a single “allow” response.
The demonstrator reconstructs the proposed action and checks evidence at the point of consequence. The goal is not to make the agent more persuasive; it is to make authority more verifiable.
Prove control of the declared software-agent identity, identify the accountable principal and verify delegation.
Check mandate, scope, value, counterparty, policy, evidence and required human authority.
Bind current authority to one canonical action digest, freshness window and one-use decision boundary.
Keep the policy decision and any later simulated execution evidence separate and accountable.
Before an agent can request a GatePass, the demonstrator can require evidence that the requester controls a registered software-agent identity, names an accountable principal and carries a valid signed delegation for the current request.
A stored approval status is not treated as proof that a real authorised person approved the exact action. The demonstrator checks active organisational identity, authentication evidence and current authority for the relevant action and value.
GatePass is the scoped, time-bound, action-specific proof primitive at the centre of ATG. It is intended to express that the trust checks were satisfied for one exact proposed action — not that an agent is trusted forever.
Authority binds to a canonical representation of the action. Changing the amount, counterparty or other bound detail changes the action and invalidates prior proof.
Nonce and replay logic demonstrate a single-use boundary. Replayed proof is refused rather than silently accepted.
Verifier-owned time and expiry checks prevent stale proof from being treated as indefinitely valid.
Changed action, unknown or revoked key, expired proof and unresolved trust state are designed to refuse in the local demonstrator.
The gate records what it decided and why. An allow decision does not automatically claim that execution happened.
Simulated execution evidence is kept separate so “allowed” and “executed” remain different facts.
ATG's Verified Human Authority roadmap keeps algorithm identifiers, key IDs, signature suites and verification policy replaceable. Current passkeys/WebAuthn are valuable phishing-resistant authentication mechanisms but are not described as automatically post-quantum secure.
Crypto-agile and post-quantum-ready in design direction — not presently claimed quantum-safe.
That distinction matters. The public repository documents future-readiness while preserving accurate claims about what is implemented today.
The corporate presentation does not broaden the implementation claims. The public build remains a local reference demonstrator and reviewer environment.
No live agent interception, production middleware, cloud enforcement service or distributed replay store is claimed.
No real payment, settlement, wallet, banking or checkout authority is executed by the demonstrator.
No legal, regulatory, compliance, security, procurement or safety certification is claimed.
Reproducible criticism, integration questions and bounded design-partner discussions are welcome.