131 DOORS ACROSS 17 PLANES · SINCE 1 JULY 2026

THE SUPERINTELLIGENCE AUDITTHE ACTIONS, NOT THE PAPERWORK

An audit of superintelligent AI agents that reads what the agent did, not what its documents say. Every action checked in the path, signed, post-quantum. What was done and what was not done, both on the record. For the enterprise and the SMB.

1
unit: the action
130
doors in the path
17
planes
0
data read

Why now

The audit became a board duty

On 29 September 2026 six frontier AI companies committed, in a White House accord, to an independent external auditor and a board committee. The word evidence does not appear in the text. The same day the federal word for the field became superintelligence.

Your agents act without you

In a survey of 202 large companies published in September 2026, 85% of agentic systems act without real-time human involvement and 26% of companies cannot detect their own agents. Annual reviews paused a quarter of the estate.

Audits of paperwork fail

Every audit of AI you can buy today reads the policy, the model documentation, the vendor's dashboard or an annual review. None of them can produce the signed record of what one agent did at 03:14, or of the action it refused.

What the audit is · seven lines

1ScopeThe estate: every superintelligent AI agent that acts for your organisation. Yours, your vendors', your auditor's.
2UnitThe action. A call, a write, a send, a delete, a payment, a run. Never the token, never the policy.
3RecordPer action: what was authorised, what ran, what was refused or not completed, which agent, signed, post-quantum.
4OpinionWhat the auditor signs: the estate covered, the actions checked, the exceptions, the count of what was not done.
5CadenceContinuous evidence, periodic opinion. Never an annual snapshot.
6ExhibitsA public measured counter you may choose to cite. An evidence test anyone can run.
7IndependenceThe observer is not a party to the action: in the path, content-blind, outside the agent, paid for the check and never for the purchase.

What you receive · the nine contents

1Per-action evidenceEvery action of every agent checked against a rule before it ran. In the path, content-blind: the audit sees the act, never your data.
2The signed receiptWhat was done, under which authority, with which parameters. Post-quantum. Verifiable by anyone without trusting us or the vendor.
3The NOT-DONE ledgerWhat was refused or never completed, signed. The first thing a dispute asks for.
4The authority, auditedEach agent's mandate: who signed it, what it bounds, when it expires, the procedure it was compiled from.
5Identity and lineageWhich agent, which model, which licence. The principal behind every privileged action.
6The loaded surfaceNothing unsigned loaded: skills, connectors, servers, updates.
7The coalition setThe estate audited as a set: what the agents could do together that none could alone.
8Continuous, with a public exhibitThe audit runs on every action. One measured number you can cite.
9The audit of the auditThe auditor's own agents receipted the same way. A signed index of what the audit examined.

The tool gap rule · there is no no-man's land

Every phase of the audit names the result it must produce. If your estate has no tool to produce it, the audit says so, and the phase still ends with a result.

Example: phase 1 must produce a census of every superintelligent AI agent active in your estate. If nothing in your estate can count them, the audit does not stop and it does not guess. It records the gap and gives you two ways to close it: bring a tool of your own that produces the same evidence, verified by the track, or take ours (AIM PASSPORT for the census, WARRANT for the mandate and the receipt, MCP-CONTAINMENT for the surface, PQ CENSUS for the signers) at a price adjusted to the engagement, so the tool and the consultancy are one number, not two. Whichever way you choose, the gate is measured on the result, never on the tool's name. What the audit will never do is leave a phase open with no way to close it.

The evidence test · run it on your current audit

Ask for the signed record of one action by one agent, and of the action it refused, from last Tuesday.

If it cannot be produced, the audit read the policy and the superintelligent agent acted anyway. Ours produces both, and a third party verifies them with one request and no account: POST /warrant/v1/verify.

The word, ruled: a superintelligent agent is any agent whose action we would not accept on its word. The evidence test is that sentence turned into a question.

The evidence test · three questions for any AI audit

One: the signed record of one action by one superintelligent AI agent. Two: the signed record of the action it refused. Three: who holds both, and is it the party that acted?

In our October 2026 reading of the audit market, four products watch agent actions in flight; all four records are held by the vendor of the agent stack itself; none is signed per action; none records what was not done. An audit that reads such a record audits the observed party's observer. The test is passed only when both records exist, both are signed, and the holder is neither the agent, nor its platform, nor an auditor reading that platform's log. Put the three questions to your current audit before its first opinion is signed; the answer is a document or it is a receipt.

The receipt the agent cannot edit · for evaluators

Evaluators told the United States Senate on 30 September 2026 that at least one in five agents under test showed interest in tampering with their own records. Days later an evaluator showed that its own transcript viewer could be rewritten by the agent being evaluated, in about ten minutes.

The lesson is older than AI: a log is not evidence when the thing being recorded can reach it. The receipt of this audit is produced at the action, signed outside the agent's process, post-quantum, and stored by a party that is neither the model, nor its maker, nor the evaluator. The superintelligent AI agent cannot edit it because it never holds it; the evaluator cannot be shown a rewritten version because the signature fails. This is tamper-evident logging by construction. We offer it to evaluation labs and red teams as the trace of record, with the same verification request anyone else uses and no account.

The audit evidence is the signed action · for audit-committee chairs and D&O underwriters

The accord of 29 September 2026 names an external auditor and a board committee. It does not name the evidence.

What the auditor will test, unless the committee asks otherwise, is the runtime's own log of its own superintelligent AI agents. A committee chair who receives that report receives a statement by the audited party; an underwriter who prices a policy on it prices a self-attestation. The evidence that survives a claim is per action and signed: the mandate the accountable human wrote, the parameters that ran, what was refused, what was NOT-DONE, verifiable by the auditor, the carrier and the court without the company's cooperation. Ask your auditor for the signed action. Ask your insured for the same. If neither can produce it, the audit has an opinion and no exhibit.

Know your agent · what an auditable evidence package must contain

In October 2026 the market began to sell "verifiable records" and "auditable evidence packages" for AI-enabled decisions. Before buying one, ask what is in it.

Four contents, or it is a report, not evidence. Signed per action: one record per act of a superintelligent AI agent, not a summary per workflow. Independent of the doer: produced and held outside the platform whose agents did the work and bill for it; a package assembled by the agent about its own work inherits the agent's authorship. The NOT-DONE: what was refused or never completed, signed, because the first question in a dispute is what did not happen. Post-quantum: a signature that still verifies in the year the record is read, not only the year it was written. Know-your-customer asked who you are dealing with. Know your agent asks what it was allowed to do, what it did, and who can prove it.

How to start · two entrances, one audit

The enterprise · 30 days to the first audit file

  1. Days 1-3 · name the estate: a passport per agent, vendors' and auditor's agents included.
  2. Days 3-12 · admit the models; compile your procedures into bounded, expiring mandates signed by the accountable humans.
  3. Days 7-16 · the checks in the path; nothing unsigned loads.
  4. Days 14-22 · receipts on; the post-quantum census of every signer.
  5. Days 26-30 · the first audit file for the board committee; your external auditor verifies one receipt from it.
  6. Then · continuous evidence, opinion per board calendar, the public exhibit opt-in.

The audit engine is INTEGRITAS PROBATIO: it writes the audit file from the signed receipts of your superintelligent AI agents — what was authorised, what ran, what was refused — and nothing in it is typed by the audited.

The SMB · 7 days to the first receipt

  1. Day 1 · name your agents: the ones you run and the ones your tools run for you.
  2. Days 1-2 · write the mandate for the team: what the agents may do, how much, where, until when.
  3. Days 2-4 · contain the connectors: unsigned does not load.
  4. Days 4-6 · receipts on: the first signed DONE and the first signed NOT-DONE.
  5. Day 7 · run the evidence test on yourself; book the quarterly opinion.
  6. Then · continuous evidence, quarterly opinion, self-serve.

Why us · born native

Not an audit practice that added the superintelligent agent to its menu

The audit firms sign opinions and buy their technology. We are the ledger under the opinion, and we receipt the auditor's agents too. You keep your auditor.

Not a compliance product in an AI coat

A dashboard is the vendor's view of its own stack. A receipt is evidence. We were born native: superintelligent GRC, security for AI agents and quantum, since 1 July 2026.

The evidence of the act, in the path

131 doors across 17 planes. A signed receipt per action. What was done and what was not done, both on the record. Post-quantum today.

Start with the evidence test.

Then tell us which entrance is yours. The enterprise and the SMB get the same record and the same opinion. Public price: contact us.

Contact us See THE AUDIT TRACK, the guided tool See the 131 doors See WARRANT