Regulations
What NOA’s receipts provide toward EU AI Act, SOC 2, and NIST AI RMF — and what’s still on the roadmap.
A capability-to-requirement map, not a compliance claim. Every row below cites the exact checked-in file a “shipped” claim comes from — verify it yourself instead of taking our word for it — and every roadmap item is labeled plainly as not shipped.
Not a certification
This page maps specific NOA capabilities to specific framework requirements — it does not mean NOA is “compliant,” “certified,” or “audit-ready” under EU AI Act, SOC 2, or NIST AI RMF. Every one of these frameworks requires organizational controls — policies, training, incident response, vendor management — beyond any single tool. Consult your own compliance team before relying on this for a regulatory submission or an audit.
Record-keeping, human-oversight, and traceability expectations for high-risk AI systems — the requirements a governance layer can most directly speak to.
Record-keeping — automatically generated logs of a high-risk AI system’s operation, kept across its lifecycle.
ShippedNOA provides — Every gated action emits a signed, hash-chained receipt the moment it is decided — verdict, risk class, policy mode, and a hash of the parameters, never the raw parameters themselves.
Human oversight — high-risk AI system decisions must remain open to human review and intervention, not run unattended.
ShippedNOA provides — A risky action is held as a signed DEFERRED receipt until a human approves it — producing a DEFERRED → ALLOWED → EXECUTED chain, not a silent auto-run.
Traceability — the ability to trace a system’s decisions back through an unbroken record.
ShippedNOA provides — Each receipt link-hashes to the one before it (chain.prevHash → chain.hash); altering or removing a past record breaks every hash that follows it.
Retained, exportable records for a regulator or auditor to inspect after the fact.
RoadmapNOA provides — Structured export tooling (the kind a regulator inspection would use) is not shipped yet — it’s listed as a paid-tier / roadmap item on this site’s own pricing FAQ, not claimed as available today.
Trust-service criteria around audit-trail integrity and independently checkable evidence — and the one thing tooling alone cannot produce: the auditor’s report itself.
Audit-trail integrity — evidence that activity logs cannot be silently altered without detection.
ShippedNOA provides — Signed, hash-chained receipts: altering any past receipt breaks the hash chain from that point forward, and every receipt is Ed25519-signed.
Independently verifiable evidence — an auditor should be able to check evidence without trusting the vendor’s own systems.
ShippedNOA provides — Verification runs fully offline against a public keyring — no NOA account, no network call — using the open-source (Apache-2.0) verifier.
A completed SOC 2 report (Type I or Type II) — an independent auditor’s attestation.
RoadmapNOA provides — Not produced by this tooling and not claimed here. SOC 2 readiness materials are a paid-tier / roadmap item; NOA does not assert SOC 2 compliance or certification today.
The four core functions — GOVERN, MAP, MEASURE, MANAGE — and which shipped capability supports each one.
GOVERN — document the policies and decision criteria an organization uses to manage AI risk.
ShippedNOA provides — Every receipt records the policy verdict (ALLOWED / BLOCKED / DEFERRED / EXECUTED / …), the mode the gate ran in, and which rule produced the decision — a durable record of the governing policy itself, not just the outcome.
MAP — categorize the risk of an AI system’s components and context before acting.
ShippedNOA provides — Every gated action carries an explicit risk-class label — LOW, MEDIUM, HIGH, CRITICAL, or IRREVERSIBLE — set before the action runs.
MEASURE — test the system, including adversarial testing, and track results over time.
ShippedNOA provides — A six-class adversarial Ed25519 test-vector suite (cofactor-malleated signatures, small-order keys, out-of-range scalars, non-canonical encodings, loose base64) plus TS↔Python cross-implementation conformance — both re-run automatically, not a one-time claim.
MANAGE — respond to identified risk, including defined escalation paths and bounded remediation timelines.
ShippedNOA provides — A risky action escalates to a human-approval hold; if nobody responds, the hold expires into its own distinct EXPIRED terminal state — never silently auto-allowed — bounded by a configurable TTL, not left open indefinitely.
Read the spec. Verify a receipt yourself. Then ask your compliance team.
Open-source, Apache-2.0. The receipt format, the risk-class labels, and the human-approval gate are all public — nothing above depends on trusting us.