The Authority Center

Why was the action allowed or denied?

One page. Six steps. Thirty seconds. Every autonomous decision an AI agent makes leaves a trail you can follow without trusting anyone — including us.

Agent attempted to approve a $1,800,000 cyber insurance claim.
Authority limit: $500,000.
Decision: DENIED.
Proof generated and independently verified.
The Six Steps
From request to provable verdict.
1 · Request

What did the agent try to do?

Agent agent.claims.alpha submitted a claim approval for $1,800,000 against policy CYB-2026-0489 at 2026-06-13 14:32:17 UTC.

2 · Authority

What was the agent allowed to do?

At registration the agent committed to a scope envelope authorizing claim approvals up to $500,000 in US-NY jurisdiction, for cyber and property action classes only. That envelope was signed by the carrier and is part of the agent's permanent identity.

3 · Policy

What rule did the substrate evaluate?

Carrier policy CLM-2026-Q3: claims above $1,000,000 require a Tier-3 escalation. The request's amount of $1,800,000 exceeded both the agent's authority cap and the escalation threshold.

4 · Decision

DENIED on DecisionAmount.

The request crossed the agent's authority by $1,300,000. The substrate refused to sign an execution receipt. The agent could not act: the action never reached the policy system.

5 · Receipt

A portable proof was emitted.

The denial produced a .h33pqv.json artifact carrying the request, the committed authority, the failed scope axes, the governance state at decision time, and three independent signatures. Anyone with the file can re-derive every claim above without trusting H33.

6 · Independent Verification

The carrier's auditor reproduced the verdict.

Audit Firm Omega, running its own copy of the verifier, recomputed every hash and re-checked every signature against the registered identities. Same artifact, same verdict: DENIED · DecisionAmount. Independent of H33, the carrier, and the agent.

Under the hood · cryptographic detail

The artifact

A single self-contained .h33pqv.json file carrying the original denial attestation, the committed scope envelope, the agent's authority request, the substrate's canonical replay frame, the Q-Key boundary, and the issuer's triple-family pubkeys plus signature.

Load-bearing identifiers

artifact schema
urn:h33:qsign:nap:v0.1
artifact_binding_hash
SHA3-384 over canonical(schema, human_summary, issuer, bindings, evidence)
scope_violation_hash
SHA3-384 over (agent_id, request_hash, committed_scope_hash, committed_gsrh, scope_check, at_unix_ms)
frame_fingerprint
SHA3-384 over the substrate's canonical replay frame at decision time

Signatures

Every signing event uses a triple-family bundle of post-quantum signatures under a 3-of-3 AND-gate — ML-DSA, FN-DSA/FALCON, and SLH-DSA must all validate:

ML-DSA-87
NIST L5 lattice (CRYSTALS-Dilithium)
SLH-DSA-256s
SPHINCS+ hash-based, NIST L5
FALCON-1024
NTRU-lattice signature

No classical schemes anywhere in the chain. Acceptance requires all three families to validate (3-of-3); a bundle where any single family fails is non-conformant, and forging a conformant bundle would require defeating all three families simultaneously.

Independent verification

The auditor runs qsign-nap-verify --registry replica.jsonl artifact.h33pqv.json and gets a structured exit code: 0 valid · 1 invalid proof · 2 malformed · 3 unsupported schema. Stdout lists every cross-check the verifier performed in alphabetical order so the output diffs cleanly across runs.

What this artifact does NOT carry

No customer data. No PII. No model weights. No proprietary policy text. Only the hashes of those things — enough to prove the substrate evaluated the right inputs, never enough to reconstruct them.

Drop the artifact yourself

Drag-and-drop verifier for NAP artifacts (v0.2) · HATS Phase 6 demo · H33-Root substrate documentation

Built for the seven people who ask the same question
Fortune 500 executives CISOs Auditors Insurance carriers Compliance officers Investors Regulators
What this page is

This is a supporting page of Authority Center, whose owned verb is PRESERVE: it holds the durable, replayable record of who could do what — the authority state — under limits, delegation, and escalation. This walk-through makes one preserved authority record legible end to end. Authority Center does not govern the decision (that is Agent-008), does not render the verdict (that is Verification), does not record the operations (that is HATS), and does not produce the portable evidence (that is H33-74). Preserving is none of those.

Why it exists

A valid identity with no live authority can do nothing, and an expired delegation revokes authority without touching identity. So “who was allowed” has to be preserved as its own answerable record — recomputed by replay over signed events, not read from a mutable database row.

When NOT this page

If you want the decision itself governed, go to Agent-008. If you want a third party to render the verdict independently, go to Verification. If you want the operations recorded over time, that is HATS. Authority Center answers only: who could do what, when, and what changed.

When to use this: use it when you need to answer “who could do what, and under what limits, delegation, and escalation” at a moment in time — a durable, replayable authority record that stays answerable across time, not read from a mutable database row.

Where to go next

Preserve/verify who holds which authority: Authority Center. Govern the decision made under that authority: Agent-008. Get the independent verdict: Verification. See the same authority across four tenses: present · past · transition · future.

Questions people ask
What does Authority Center preserve?

Authority state — the durable, replayable record of who could do what at a moment, under what limits, escalation, and delegation. This page shows one such record end to end. The dollar figures are labeled sample data, not a live customer record.

Does it govern the decision?

No — governing the decision is Agent-008. Authority Center preserves the authority state the decision was made under so it stays answerable later. Preserving is not governing.

How is it different from Agent-008 and Verification?

Agent-008 governs the decision; Verification renders the independent verdict; Authority Center preserves the authority both rest on. Steps 1–5 above are the preserved record; step 6 is the Verification path.

Is this a live decision or a sample?

The walk-through uses labeled sample values to teach the six-step chain. The mechanism — the artifact, the triple-family post-quantum signatures under a 3-of-3 AND-gate, and the independent verifier — is real. Run the authority question against your own environment via the evaluation path below.

Next step

See exactly what happened

The auditor view of this same decision — reconstructed millisecond by millisecond from inputs you can verify yourself.

Open Replay