The engine
Proof your auditor checks on their own — without seeing your data.
What’s inside an audit pack, how rules are versioned, and exactly what gets checked when someone verifies a result.
The audit pack
What your auditor receives.
Instead of a folder of exports, your auditor gets a self-contained pack of evidence that can be checked without us.
- Signed results
- A pass or fail for each control, signed at the moment it was checked.
- Rule fingerprints
- Each result carries the fingerprint of the exact rule version it was checked against.
- The whole period
- Every check across the period under review, in order — not a sample chosen afterwards.
- Proofs for sensitive facts
- Where the underlying data is sensitive, a zero-knowledge proof that the check passed — not the record itself.
- Manual sign-offs
- Policies, training and reviews, recorded with the owner’s sign-off and signed like everything else.
- Signed resultsPass or fail per control
- Rule fingerprintsThe exact rule version for each result
- The whole periodEvery check, in order
- Proofs for sensitive factsZero-knowledge, not records
- Manual sign-offsPolicies, training and reviews
Independent verification
What gets checked when someone verifies.
An auditor, examiner or customer can run these checks themselves. None of them needs access to ZTZK or to your systems.
- AuthenticityThe signature matches the published public key, so the result came from where it claims to.
- IntegrityThe result is byte-for-byte what was signed. Change anything and the check fails.
- Rule versionThe rule fingerprint matches the rule that was in force, so no one can swap in a friendlier rule later.
- Place in the recordThe result sits in an append-only ledger, so it can’t have been added, removed or reordered after the fact.
- Sensitive factsZero-knowledge proofs confirm the claim holds without revealing the data behind it.
Under the hood
Why the evidence holds.
For the technical co-signer: the cryptography is the reason to believe, and all of it is independently checkable.
Post-quantum signatures
Results are signed with ML-DSA-65, the NIST-standardized post-quantum signature scheme — built so the record holds up after today's encryption doesn't.
Zero-knowledge proofs
STARK-based zero-knowledge proofs let a verifier confirm a result without seeing the data it was computed from.
Append-only ledger
Every signed result is added to a ledger that can be extended but not rewritten, so the history can't be quietly edited.
Runs where your data lives
Checks run inside your own infrastructure. Your data stays in your systems — only the proofs leave.
Versioned rules
Rules your compliance team can read.
Every control is written as a short, readable contract — what must be true, what is forbidden. Compliance can review it; engineering can version it. Each version gets a unique fingerprint. Change a rule and the fingerprint changes — while results checked under the old version still verify against it, so history never has to be rewritten.
What code can’t prove — a policy, a training record, a supplier review — is captured with the owner’s sign-off and signed to the same ledger.
GUARANTEE mfa_enabled FOR admin_routes
FORBID plaintext_secret IN repository
ENSURE access_reviewed WITHIN intervalOne engine · two audit stories
The same proof, beyond compliance.
The engine that proves your compliance controls also enforces AI policy on every call — and proves what the model did, on the same ledger, without exposing the model or its data. AI governance is in early access today.
Request access
See the engine on your controls.
We're onboarding a small number of early teams. Tell us what you need to prove — and to whom — and we'll walk you through it.
We use your email only to reply to your request. Privacy policy
✓ Request received. We'll be in touch.