NIST AI Risk Crosswalk

Version: 1.0.1
Date: 2026-07-16
Status: Informative

Purpose

Map OAIES artifacts to NIST AI Risk Management Framework 1.0 functions and the NIST AI 600-1 Generative AI Profile.

Why

The AI RMF organizes risk work as GOVERN, MAP, MEASURE, and MANAGE. NIST AI 600-1 applies that structure to generative-AI risks; it does not create a certification program or prescribe one implementation.

AI RMF crosswalk

NIST AI RMF function/category	OAIES evidence	Relationship	Evidence gap to close
GOVERN 1–2: policies, accountability, roles	Control Catalog; AI System Manifest owner and executive; Statement of Applicability	`partial`	Workforce culture, training, and organization-wide policy effectiveness
GOVERN 3: workforce diversity, culture, incentives	No direct machine contract	`no-mapping`	HR, competence, participation, and incentive records
GOVERN 4: organizational risk culture	Risk Register; Exception; Incident	`partial`	Enterprise risk appetite operation and escalation behavior
GOVERN 5: engagement with relevant AI actors	Manifest affected parties; Model Record; MCP Server Record	`partial`	Consultation and communication outcomes
GOVERN 6: third-party and supply-chain risk	Model Record source/license/data governance; MCP Server Record; Evidence Bundle	`partial`	Contracts, supplier monitoring, and exit plans
MAP 1: context and risk framing	Manifest purpose, uses, affected parties, P/A/I classifications, jurisdictions	`evidence-supports`	Broader social context may require external research
MAP 2: system categorization and boundaries	Manifest components, lifecycle, boundary, human oversight	`evidence-supports`	Validate deployed configuration against the manifest
MAP 3: benefits and costs	Manifest purpose; Risk Register	`partial`	Quantified benefit and opportunity-cost analysis
MAP 4: risks and impacts	Risk Register scenarios, affected parties/assets, inherent/residual ratings	`evidence-supports`	Stakeholder impact evidence and uncertainty analysis
MAP 5: impact characterization	Risk Register; Evaluation Result	`partial`	Aggregate, long-term, and systemic impact methods
MEASURE 1: measurement approaches	Evaluation Result suite, dataset, metric, threshold, confidence interval	`evidence-supports`	Method validity and independent review
MEASURE 2: trustworthy characteristics	Evaluation Result; Context Manifest; Prompt/Model records	`partial`	Coverage must be risk-appropriate and system-level
MEASURE 3: tracking and feedback	Evaluation Result; Incident; Evidence Bundle	`partial`	Production feedback channels and affected-party access
MEASURE 4: expert review and contextualization	Evaluation executor/evidence; approvals	`partial`	Reviewer qualifications and independence
MANAGE 1: prioritize risks	Risk Register level, disposition, owner	`evidence-supports`	Alignment with enterprise appetite requires separate proof
MANAGE 2: plan and implement treatment	Risk treatment actions/control refs; Statement of Applicability	`partial`	Operational-effectiveness evidence remains required
MANAGE 3: manage third-party risk	Model/MCP records; Evidence Bundle	`partial`	Contractual remedies and continuous supplier assurance
MANAGE 4: monitor, respond, recover, communicate	Evaluation Result; Incident; Exception	`partial`	Exercised response plans and communications evidence

NIST AI 600-1 Generative AI Profile crosswalk

NIST AI 600-1 identifies generative-AI risks and recommended actions. The profile’s risk names should be preserved in the Risk Register rather than replaced with OAIES labels.

GAI risk area	Required OAIES implementation evidence	Relationship
CBRN information or capabilities	Prohibited uses, authority boundaries, domain red-team evaluations, incident playbook	`partial`
Confabulation	Groundedness/factuality evaluations, user disclosure, human verification gates	`evidence-supports`
Dangerous, violent, or hateful content	Safety evaluations, prompt release security review, monitoring and incident evidence	`evidence-supports`
Data privacy	Context source classification/authorization, model data governance, retention and redaction evidence	`partial`
Environmental impacts	Supplier/model energy and resource evidence attached to Model Record	`partial`
Harmful bias and homogenization	Stakeholder-specific evaluation slices, impact risks, treatment decisions	`evidence-supports`
Human-AI configuration	Human oversight mode, intervention point, authority limits, UX evaluation	`evidence-supports`
Information integrity	Provenance, content integrity, misinformation evaluations, incident response	`partial`
Information security	Tool/MCP authorization, context isolation, adversarial evaluation, incident handling	`partial`
Intellectual property	Model source/license, dataset rights evidence, output-use controls	`partial`
Obscene, degrading, or abusive content	Safety evaluations, prohibited uses, reporting and response	`evidence-supports`
Value-chain and component integration	Versioned component records, digests, supplier evidence, dependency inventory	`partial`

How

Use every AI RMF function; do not start at MEASURE.
Register applicable NIST AI 600-1 risks using the source risk name and local scenario.
Bind each risk to system components, affected stakeholders, metrics, treatments, evidence, and owner.
Define pre-deployment gates and production indicators separately.
Record uncertainty, test limitations, and residual-risk decisions.
Re-run MAP and MEASURE when use, model, data, tool, autonomy, or deployment context changes.

Equivalence limits

The NIST AI RMF and NIST AI 600-1 are voluntary risk-management resources unless another authority makes them mandatory. NIST does not certify implementations against them. OAIES mappings indicate evidence alignment, not NIST compliance, conformity, endorsement, or complete implementation.

Tradeoffs

Benefit	Cost
Risk outcomes remain traceable to engineering releases	Framework tailoring requires documented judgment
Common GAI risk vocabulary improves reporting	Profile actions do not cover every domain-specific risk
Metrics become reproducible artifacts	Measurement can create false confidence without MAP context

Anti-patterns

MEASURE-only adoption: running benchmarks without governance, context mapping, or treatment.
Universal threshold: applying one quality or safety threshold across stakeholder groups and use cases.
Risk-name substitution: renaming source risks until traceability is lost.
NIST-certified claim: NIST AI RMF has no NIST certification mechanism.

Enterprise considerations

Integrate AI risks into enterprise risk taxonomy, board reporting, procurement, privacy, security, and business continuity. Preserve dissent and uncertainty in acceptance decisions. High-impact systems require independent evaluation and affected-stakeholder participation.

Authoritative sources

Sources accessed 2026-07-16.

Checklist

GOVERN, MAP, MEASURE, and MANAGE all have accountable evidence
Applicable AI 600-1 risks retain source terminology
Metrics include dataset/version, threshold, sample size, and limitations
Third-party and post-deployment risks are included
Residual risk is accepted by the correct authority
No NIST certification or endorsement claim is made

Changelog

1.0.1 — 2026-07-16

Made relationship claims evidence-specific and aligned mappings with P/A/I classifications.

1.0.0 — 2026-07-16

Added AI RMF 1.0 function/category and AI 600-1 GAI risk mappings.