Compliance Pack · NIST AI RMF 1.0
Shipping Q3 2026Your NIST AI RMF evidence, already produced.
SynOI features map directly to specific NIST AI RMF subcategories. GOVERN-1.1, MAP-2.3, MEASURE-2.6, MANAGE-4.1: each one backed by cryptographically verifiable Decision Receipts you can hand your auditor. They verify offline; no SynOI account, library, or cooperation required.
Why NIST AI RMF
The US AI compliance lingua franca.
Federal AI procurement standard
OMB M-24-10 makes AI RMF alignment effectively required for selling AI to federal agencies. Enterprise procurement RFPs increasingly cite specific subcategories.
AI-specific, not generic security
SOC 2 covers your security posture broadly. NIST AI RMF covers what your AI system actually does, decides, and learns. Different evidence, different conversation.
Voluntary but de facto
No third-party audit required to self-attest. But your enterprise customers will ask. Having auditable evidence makes that conversation 30 minutes instead of three months.
Built on substrate you already have
SynOI's Decision Receipts, HITL records, integrity attestations, and policy versioning are exactly the evidence subcategories request. We just packaged it.
What you get
Four artifacts. One workflow.
Mapping document
Every NIST AI RMF subcategory we evidence is mapped to a specific SynOI feature. ~70% of MEASURE and MANAGE subcategories directly satisfied; full GOVERN coverage; MAP subcategories partial (organizational mapping is out-of-scope for any software product).
Compliance dashboard view
Filter receipts by NIST subcategory. See coverage in real time. Drill from 'MEASURE-2.6' down to the specific receipts that constitute its evidence. Spot gaps before your auditor does.
Per-quarter audit export
One bundle: PDF for the auditor's binder, CSV for their spreadsheet, structured JSON for their tooling. Receipts organized per subcategory. Verification instructions inline.
Auditor support
On Enterprise tier: we get on a call with your auditor and walk them through the evidence. They've audited Vanta-style SOC 2 work; AI RMF is new enough that they appreciate the hand-off.
Coverage breakdown
Direct evidence by NIST function.
GOVERN
75%6 of 8 subcategories evidenced
- GOVERN-1.1: Legal requirements documented
- GOVERN-1.4: Testing + incident response
- GOVERN-2.1: Roles + responsibilities
- GOVERN-6.1: Third-party AI risks
MAP
67%5 direct + 5 supporting subcategories evidenced
- MAP-1.1: Intended purposes
- MAP-2.1: Task definitions
- MAP-2.3: TEVV considerations
- MAP-5.1: Risk likelihood
MEASURE
69%9 of 13 subcategories evidenced
- MEASURE-2.1: Test sets + metrics
- MEASURE-2.6: Trustworthiness evaluation
- MEASURE-2.7: Security + resilience
- MEASURE-2.11: Fairness + bias
MANAGE
92%11 of 12 subcategories evidenced
- MANAGE-1.2: Risk treatment prioritization
- MANAGE-2.4: Supersede / disengage
- MANAGE-3.1: Third-party monitoring
- MANAGE-4.1: Post-deployment monitoring
Subcategories not directly evidenced are organizational mapping or culture work (GOVERN-4.1 "safety-first mindset," MAP-3.1 "potential benefits examined"). No software product can satisfy those - only your humans can. SynOI gives you the substrate; your humans do the mapping. Combined, you have the full RMF.
Sample evidence for one subcategory
MEASURE-2.6: Trustworthiness
"The AI system is evaluated regularly for safety, security, accuracy, reliability, transparency, accountability, explainability, and fairness." Every property satisfied by a specific SynOI feature:
HITL gates on high-risk action classes; cryptographic record of approver identity per decision.
Integrity heartbeat catches tampered gateway code; flagged installs marked in receipts.
Model ID and complexity score stamped on each receipt; correlate outcomes per model.
Cache hit rate, p95 decision latency, error rate: all queryable per-period.
Receipts are publicly verifiable via npx @synoi/verify; auditor needs no SynOI cooperation.
Hybrid Ed25519 + ML-DSA-65 signatures bind every action to the gateway that performed it. Non-repudiable.
action_desc field + linked LLM rationale CDRO. Why the AI chose this action, recorded with it.
Per-subject queryable receipts feed standard bias-audit pipelines (NYC LL144-compatible exports).
On every receipt
compliance_tags: pre-mapped, deterministic, signed.
Every Decision Receipt carries a sorted JSON array of subcategory tags derived deterministically from the receipt’s action_class, decision, risk_level, action_type, and HITL state. The same inputs always produce the same tags. Filter by framework at audit time without re-deriving from raw fields.
{
"receipt_id": "rcpt-1779234155-9q3kz1xv",
"decision": "allow",
"action_class": "B",
"risk_level": "low",
"action_type": "AnthropicMessages",
"signature": "ed25519:...",
"compliance_tags": [
"EU.Article12",
"EU.Article16",
"EU.Article26",
"ISO42001.Clause-8.1",
"ISO42001.Clause-9.1",
"NIST.MANAGE-1.2",
"NIST.MANAGE-4.1",
"NIST.MAP-2.1",
"NIST.MEASURE-2.6",
"SOC2.CC7.1",
"SOC2.CC9.2"
]
}The tag set is namespaced so one receipt can satisfy multiple frameworks at once. A single HITL approval, for example, is evidence under NIST.GOVERN-1.4, EU.Article14 (human oversight), ISO42001.Annex-A.6 (operational impact), and SOC2.CC7.3(incident response) - all in the same line of JSON. Auditors filter by prefix; you don’t re-map anything.
False positives are cheap (an auditor sees one extra receipt); false negatives are expensive (a compliance gap because the receipt wasn’t filed correctly). The tagger is intentionally conservative - when in doubt, the tag is included.
How to get it
Included in your plan.
Free (self-host)
Substrate only - no Compliance Packs.
Personal $19/mo
Pick NIST AI RMF as your 1 included Pack.
Team $99/seat
Included (one of three Packs).
Enterprise
Included with auditor support + custom mappings.
Have an auditor citing NIST AI RMF?
Tell us which subcategories they're asking about. We'll show you exactly which SynOI receipts answer them.