NIST AI RMF vs ISO 42001: which one a customer or contract is asking for
The NIST AI RMF is free, voluntary and cannot be certified, while ISO/IEC 42001:2023 is a standard you purchase and that an independent certification body can audit and certify. Before you commit budget, read the clause in the customer contract or regulatory request and identify which of these two outcomes it actually asks for. A request to "align with the NIST AI RMF" calls for a documented self-assessment you can defend, while a request for "ISO 42001 certification" calls for a certificate issued by an accredited body after an audit of your AI management system. Those are different deliverables with different owners, so confusing them is the most common way teams end up preparing the wrong evidence.
The choice depends on what the buyer must be able to show. If they need a neutral, third-party attestation of an AI management system, only ISO/IEC 42001 certification can answer that. If they want evidence that your AI risk practices map to a published risk framework, the NIST AI RMF assessment is the right basis, and it can be produced with your own staff and an implementation partner. The NIST crosswalk is useful here, but it is a mapping and not proof of equivalence. NIST states that listing it does not imply endorsement and that neither resource comprehensively covers the other. The crosswalk was also drafted against the FDIS text rather than the published 2023 edition, so confirm each clause reference against your own copy of the standard before you rely on it.
Before you answer the customer, ask them three things in writing: which document they named, whether they require a certificate or a self-assessment, and which edition or version they expect. Ask your auditor or certification body whether they will audit against ISO/IEC 42001 for your scope, and confirm current NIST revision status, since NIST says AI RMF 1.0 is being revised and has not published a new version. The cost of ISO/IEC 42001 includes the standard itself, audit fees, and internal effort, while the NIST framework and its Playbook are free, with the work of documentation and assessment falling on your team. Get the deliverable named in the contract before you start.
How the two frameworks differ on status, structure and cost
This table compares the NIST AI RMF and ISO/IEC 42001 on status, structure, document cost and who assesses them. Read each row as a question an assessor, customer or regulator would ask, and treat the cost row as the price of the standard text only, not the price of implementing or proving either framework.
| Dimension | NIST AI RMF 1.0 | ISO/IEC 42001:2023 |
|---|---|---|
| Status | Voluntary. NIST says neither private nor public organizations are required to use it. | International standard that organizations can implement, and that a certification body can certify against. |
| Certification | No certification, accreditation or conformity assessment program exists for the AI RMF. | Certifiable. Independent certification bodies audit an AI management system against its requirements. |
| Structure | AI RMF Core: 4 functions, 19 categories and 72 subcategories of outcomes that users may select. | Management system requirements in clauses 4 to 10, with Annex A controls and Annex B implementation guidance. |
| How it is applied | Users choose which outcomes to apply and in what order. NIST says it is not a checklist. | Requirements must be met across the whole AI management system, and auditors test them as written. |
| Document cost | NIST AI 100-1, NIST AI 600-1 and the Playbook are free to download. | CHF 225 for a single-user copy of ISO/IEC 42001:2023 on the IEC Webstore, as of October 2026. |
| Who assesses | No assessment program. Organizations self-assess, and any independent review is arranged by the organization itself. | Certification bodies audit against ISO/IEC 42001, and ISO/IEC 42006:2025 (CHF 159 as a standards document) sets criteria for those bodies. |
| Revision status | Being revised under the White House AI Action Plan, with no revised version published as of October 2026. | Edition 1.0 published December 2023, with no amendments listed on the IEC Webstore as of October 2026. |
| Confirm before relying on it | The current NIST publication status and any revised version before citing a version number. | The current edition and amendments with the IEC, and certification body scope with the body itself. |
What the Microsoft crosswalk maps and where the mapping stops
NIST lists a crosswalk provided by Microsoft that maps each AI RMF subcategory to clauses 4 to 10 and Annex B implementation guidance of ISO/IEC FDIS 42001. The mapping is a reading aid, not a substitute for either document. For example, GOVERN 1.1 (legal and regulatory requirements) is mapped to clauses 4.1 and 6.2 and to Annex B items B.2.2 and B.2.4. Use the crosswalk to find where a 42001 clause already covers work you planned under an AI RMF subcategory, then check the wording of both sources yourself before you cite one against the other in a customer questionnaire or contract.
Two limits matter before anyone relies on it. First, the crosswalk was drafted against the FDIS (final draft) text of ISO/IEC 42001, not the published ISO/IEC 42001:2023 edition, according to its title. Confirm clause numbers and Annex B item references against the published edition you purchased before you map evidence to them. Second, NIST states that listing a crosswalk does not imply its endorsement of the mapped resource, nor that either resource comprehensively covers the other. An AI RMF subcategory with no matching clause is not automatically a gap in a 42001 program, and a clause with no matching subcategory may still be required for certification.
NIST's crosswalk page also lists INCITS/AI crosswalks from the AI RMF to ISO/IEC 23894 (revised) and to ISO/IEC 42005, both dated August 14, 2025. NIST's AI RMF Roadmap names alignment with international standards and production of crosswalks as a top priority, which suggests more mappings will appear, and AI 100-1 itself is under revision. Treat any crosswalk as dated to its own version. Before you rely on one, record the document version, its date and the standard edition it was drafted against, and ask the team that maintains your compliance evidence to re-check the mapping when either side changes.
Five limits to check before using a crosswalk as evidence
The NIST crosswalk page lists a mapping from AI RMF subcategories to ISO/IEC 42001 clauses, and it is a useful starting point. Each limit below affects whether a mapped item can be cited as evidence, so check them before a mapping goes into a customer response or an audit file.
Drafted against the FDIS
The Microsoft crosswalk, as its title indicates, was drafted against the FDIS (final draft) text of ISO/IEC 42001 rather than the published 2023 edition. Clause numbers, annex items and wording can differ between a draft and the final publication. Before citing a clause reference, open the published edition you hold and confirm that the clause says what the crosswalk assumes it says.
No NIST endorsement of the mapping
NIST's crosswalk page states that listing a crosswalk does not imply NIST endorsement of the resource the AI RMF is mapped to. The mapping is a third-party document that NIST hosts for reference. Do not describe a mapped control as NIST-approved or as meeting a NIST requirement, because the AI RMF is voluntary and NIST does not validate any organization's use of it.
No claim of full coverage
NIST states that neither resource comprehensively covers the other. A subcategory with a clause reference may still have gaps against the standard's requirements, and an Annex B item may guide implementation without matching an AI RMF outcome exactly. Ask your certification body or auditor which requirements the mapping does not reach, and document those gaps in your own management system records.
Numbering may change in revision
NIST says AI RMF 1.0 is being revised as part of the White House AI Action Plan and has not published a revised version. If the subcategory identifiers change, a crosswalk keyed to the current GOVERN, MAP, MEASURE and MANAGE numbering will need updating. Record the AI RMF version you used in every evidence file, and check NIST's AI RMF pages for the revised version before you rely on any identifier in a contract.
Playbook last updated August 2023
The NIST Playbook's audit log shows its last content update in August 2023, and AIRC says it will be updated after the AI RMF is revised. Suggested actions and documentation prompts linked to a subcategory may therefore predate the current version of the standard. Treat Playbook text as background, and confirm each mapped item against the ISO/IEC 42001 text you hold and against your own environment.
Who assesses each framework and what the organization still owns
The NIST AI RMF has no certification program. NIST states that it does not validate or endorse any organization's use of the framework, so there is no certifying body to hire. The organization decides how to assess itself, and it decides whether to bring in an outside reviewer. Any report that describes AI RMF alignment is an internal assessment or an independent review that the organization commissioned and scoped. Its value depends on the reviewer's independence and the depth of the evidence behind it.
- Self-assessment: Teams compare their current practices with the subcategories they select and record gaps, owners and target dates.
- Independent review: The organization chooses a reviewer, such as internal audit or an outside firm, and agrees the scope and deliverables in writing before work starts.
- Ownership of conclusions: The organization signs its own statements about AI RMF use. Nobody outside the organization certifies them.
- Evidence: Policies, AI inventories, test results and monitoring records must exist in the organization's own systems.
- Vendor claims: A vendor saying it is 'AI RMF certified' is making a claim NIST does not support.
Ask any reviewer to name the subcategories tested, the evidence reviewed and their independence from the team that built the system.
ISO/IEC 42001 is a certifiable management system standard. An accredited certification body audits the organization against its requirements and issues a certificate if the audit supports it. The certification body applies ISO/IEC 42006, which sets requirements for bodies that audit and certify AI management systems. Consultants and implementation partners cannot issue the certificate, even when they helped build the management system.
- Certification body: An independent body accredited for AI management system audits performs the certification audit.
- Two-stage audit: Auditors typically review documentation first, then test whether the management system operates in practice. Confirm the stages and their timing with the body you choose.
- Surveillance: Certificates depend on periodic surveillance audits, so budget for recurring assessment work.
- Accreditation: Ask whether the body's accreditation covers AI management systems under ISO/IEC 42001 before you sign.
- Scope statement: The certificate covers only the scope the organization defines, so confirm which AI systems and business units it includes.
Check the certification body's accreditation status with the accreditation body in its country before you agree on a scope.
Under both frameworks, the organization owns its controls. It sets the AI policy, names accountable owners, approves risk decisions and operates the systems. An implementation partner such as DeployOpen can plan, build, configure and document the systems that produce evidence inside the organization's own infrastructure. The partner prepares that evidence. It does not issue audit reports, attestations or certificates, and it does not decide whether a control is effective.
- Organization: Owns policy, risk acceptance, accountable roles, and the decision to submit to an assessment.
- Implementation partner: Prepares systems, configurations, documentation and evidence, and can support remediation work.
- Independent auditor or certification body: Issues the report or certificate, and only it can do so.
- Shared work: Staff must operate the controls after handover, so plan internal ownership for the evidence trail.
- Scope agreement: Define in writing which systems, teams and decisions the work covers.
Confirm in the statement of work who signs each deliverable and who receives the final evidence package.
Customers and regulators often ask for an independent view, but the right reviewer depends on the framework. For NIST AI RMF work, the reviewer is a choice the organization makes and can be internal or external. For ISO/IEC 42001, the reviewer must be an accredited certification body. An implementation partner can be useful at both stages, but it should not be the party that judges its own work. Separate the people who build the evidence from the people who test it.
- Internal review: Suits early gap assessments and internal readiness checks, but may not satisfy a customer asking for independence.
- External review: Gives an outside view of an AI RMF assessment, with scope and report format agreed in advance.
- Certification body selection: Compare accreditation status, auditor experience with AI systems, and fees in writing.
- Partner role: Keep implementation separate from the final assessment so findings remain credible.
- Procurement: Ask who signs the report and whether the reviewer has any stake in the outcome.
Before you engage anyone, ask each party to state in writing what it will deliver and what it will not claim.
How a team can use both frameworks without doubling the work
The sequence lets one body of evidence serve both the voluntary AI RMF outcomes and the ISO/IEC 42001 management system records, so the team builds governance once and checks the gaps against the certifiable standard only where a certificate is required.
Confirm what the contract actually requires
Start by reading the customer contract, questionnaire or regulator request and writing down the exact wording. Decide whether it asks for the NIST AI RMF (a voluntary framework with no certificate), ISO/IEC 42001 (a standard a certification body can certify), or evidence aligned to either. Legal or procurement owns the decision, and the security or AI governance lead confirms what each clause means in practice. The step is done when the team has a one-page statement of which framework and which form of proof are required, with the owner named. If certification is required, confirm the scope, timing and certification body with the certification body itself, since those depend on its own process.
Set governance and risk outcomes with the AI RMF Core
Use the Govern function as the cross-cutting base, then work through Map, Measure and Manage for the AI systems in scope. NIST says users may choose which categories and subcategories to apply and that most start with Map once Govern outcomes exist. The AI governance lead assigns owners for each outcome, records the AI system inventory, risk tolerances and accountability structure, and keeps the Playbook as a source of documentation prompts rather than a checklist. The step is done when each in-scope system has documented context, risk decisions and named owners that a reviewer can trace back to an outcome.
Map AI RMF subcategories to ISO/IEC 42001 clauses
Use the crosswalk that NIST lists on its AIRC page to find which ISO/IEC 42001 clauses and Annex B implementation items relate to each AI RMF subcategory the team has addressed. Note that the crosswalk was drafted by Microsoft against the FDIS text of the standard, not the published 2023 edition, so the team checks each mapped clause against its own copy of ISO/IEC 42001:2023. NIST notes that a crosswalk does not imply its endorsement and that neither resource fully covers the other. The step is done when a gap register lists every 42001 clause with no matching evidence in the AI RMF work, each marked as covered, partly covered or missing.
Build management system records for each gap
For each gap in the register, the team writes the management system records that ISO/IEC 42001 expects: documented policies, objectives, roles, competence records, operational procedures, internal audit and management review outputs, and corrective action logs. The internal team owns these records, and an implementation partner may help draft or structure them when the team lacks capacity. Records should be created from real operating activity rather than written retrospectively to fit a clause. The step is done when each record has an owner, a version, an approval date and a link to the AI RMF outcome it supports, so one document can answer both framework questions.
Collect Annex A and audit evidence
Gather the evidence that shows the Annex A controls and Annex B guidance are operating, such as system-level risk assessments, testing and monitoring outputs, incident and feedback logs, third-party supplier reviews and training records. Each piece of evidence is tagged to the AI RMF subcategory and the 42001 clause it supports. The security and platform teams pull logs and reports from the self-hosted tooling in the organization's own environment, and the governance lead reviews completeness. The step is done when an assessor can sample any control and find dated, attributable evidence without the team explaining it verbally.
Run an internal review before any external assessment
Before inviting outside parties, the internal audit or governance function checks the records against the gap register and the Annex A evidence set. Reviewers test whether each claim in the AI RMF self-assessment matches what the systems actually do, and whether management review has acted on the findings. The team records open issues, owners and closure evidence. Because the AI RMF is voluntary and does not offer a certificate, this review is the team's own assurance step and should be described that way in any customer communication. The step is done when open findings are either closed with evidence or accepted by a named executive with a documented rationale.
Engage a certification body if certification is required
Only when the contract or regulator requires ISO/IEC 42001 certification does the team engage an accredited certification body. The certification body, not DeployOpen or the internal team, decides whether the management system conforms and issues any certificate. The team asks the certification body which accreditation scope applies, what its audit plan and fees are, and how it treats the 2023 edition and ISO/IEC 42006:2025 for auditing and certifying AI management systems. Timing and cost depend on the certification body's process and the organization's scope, so the team confirms both in writing before the audit is booked. The step is done when the certification body has accepted the scope in writing and the evidence set has been shared under the agreed process.
Which option fits a buyer that needs evidence now
This table compares three ways to answer a customer request: a self-assessment against the AI RMF, an ISO/IEC 42001 certification, or both. Read each row against the wording of the contract or questionnaire, because the right path depends on whether the customer asks for a framework alignment statement or a certificate. The main takeaway is that the AI RMF can be answered now without a certificate, while a contract that names 42001 certification needs an accredited certification body and a longer timeline.
| Option | When it fits | What the buyer must confirm with the customer |
|---|---|---|
| AI RMF self-assessment | A questionnaire or security review asks how AI risks are governed, measured and managed, with no mention of certification | Whether an internal assessment against the AI RMF Core is acceptable, and what evidence and independence level they expect |
| ISO/IEC 42001 certification | A contract names certification to ISO/IEC 42001 or requires a certificate from an accredited body | Which certification body is acceptable, the scope of certification, and the date by which a certificate must be in place |
| Both together | The customer wants AI RMF vocabulary for risk and a certifiable management system for assurance | Whether a mapping to 42001 clauses is accepted as supporting material, and that the mapping is treated as a working aid, not an official equivalence |
| Customer wording check | Before choosing, the buyer reads the exact requirement in the contract or questionnaire | Whether the request says framework, certificate, attestation or report, since only an independent body can issue a certificate |
| Evidence needed now | A response is due before any external audit could be completed | Whether a self-assessment with documented gaps, owners and a dated plan will satisfy the request in the interim |
| Certificate from an external body | The customer will not accept self-declared alignment for AI management | Which certification body is accredited for 42001, and what audit scope and timeline it will quote in writing |
| Crosswalk use | Teams need to show where AI RMF subcategories touch 42001 clauses | That the crosswalk is treated as a researched aid, since NIST does not endorse it and it was drafted against the FDIS text |
Questions buyers ask about NIST AI RMF and ISO 42001
Can we be certified to the NIST AI RMF?
No. The NIST AI RMF is a voluntary framework, and NIST does not validate or endorse any organization's use of it, so there is no AI RMF certificate to obtain. Your organization owns its AI risk management program and can describe how it aligns with the framework's Govern, Map, Measure and Manage functions in its own documentation. If a customer or regulator wants independent confirmation, the usual route is an assessment by an outside party, and that party's report is what carries weight, not a badge. Ask any firm offering an AI RMF certification to name the body that accredits it and the document it issues, because NIST's framework materials describe no such program.
What does the ISO 42001 standard cost to buy, and what else costs money?
The standard itself is a modest one-time purchase. The IEC Webstore listed ISO/IEC 42001:2023 at CHF 225 for a single-user copy as of October 2026. The larger costs sit around it: staff time to define the AI management system scope and policies, internal audits, management review, remediation of gaps, tooling, and the fees of the certification body. Certification body fees are recurring in practice, since surveillance audits follow the initial certification and a recertification audit follows the cycle. Confirm current fees, audit days and timing directly with the certification bodies you shortlist, and ask each for a written quote based on your scope.
Who issues an ISO 42001 certificate?
An accredited certification body issues the ISO/IEC 42001 certificate after auditing your AI management system. Your organization does not certify itself, and neither does an implementation partner. ISO/IEC 42006:2025, which the IEC Webstore lists at CHF 159, adds requirements to ISO/IEC 17021-1 for bodies that audit and certify AI management systems against ISO/IEC 42001, and it can serve as a criteria document for accreditation or peer assessment. Before engaging a body, ask whether it is accredited for ISO/IEC 42001 by a recognized accreditation body, and whether its certificate scope covers the AI systems and business units you intend to include.
Can the NIST crosswalk to ISO 42001 replace an audit?
No. The crosswalk is a reference document that maps NIST AI RMF subcategories to ISO/IEC 42001 clauses 4 to 10 and Annex B implementation guidance. It helps teams reuse evidence across two frameworks, but NIST states that listing a crosswalk does not imply endorsement of the mapped resource, nor that either resource comprehensively covers the other. The Microsoft-provided crosswalk was drafted against the FDIS text of ISO/IEC 42001 rather than the published 2023 edition, so check each mapped clause against the published standard before relying on it. An auditor will test your management system and its evidence directly, regardless of any mapping.
Should we start with NIST AI RMF or ISO 42001?
Start with the question your buyer or regulator is asking. If a customer contract or regulator names the NIST AI RMF, begin there, because it is free, its functions are outcome-based, and it helps you identify where your AI risks sit. If you need a certificate that a customer can verify, ISO/IEC 42001 is the certifiable option, and the work to build its management system overlaps heavily with the Govern function. Many organizations build one governance program and then document it against both. Confirm with each customer which standard they actually accept, since some ask for the framework, some for the certificate and some for both.
What does the work need from our team?
The work needs named owners across legal, security, product, data science and risk, plus an executive sponsor who can approve AI risk decisions. Your team must supply the inventory of AI systems, their intended purposes, their third-party models and data sources, and the evidence of testing and monitoring that already exists. An implementation partner can prepare systems, controls and documentation, and an independent auditor or certification body issues any report or certificate. Ask any partner to state in writing which tasks it performs, which remain with your staff, and which belong to the auditor, so your internal effort is scoped before work starts.
How DeployOpen prepares AI management and AI RMF evidence
DeployOpen prepares the systems, controls and documentation that an organization needs for an AI risk program or an AI management system, whichever route it takes. We do not issue audit reports or certificates; an independent auditor or certification body does that work.
How we scope
- Engagements are fixed-scope, with the AI systems, business units and framework route agreed in writing before work starts.
- We confirm whether the goal is an AI RMF self-assessment, an ISO/IEC 42001 readiness effort, or both, and record that choice in the scope.
- Written scope names the deliverables, the systems in the self-hosted environment and who in the client organization owns each control.
How we build
- We set up self-hosted open-source platforms inside your own cloud accounts, data centers or air-gapped networks, with logging that can support AI system records.
- We map AI RMF subcategories or ISO/IEC 42001 requirements to the systems and procedures that produce evidence, using the NIST crosswalk only as a starting reference.
- We test the controls we build, including AI system testing and security testing where the written scope includes it, and record the results.
How we hand over
- You receive the policies, inventories, risk records and evidence files defined in the scope, organized so an assessor can trace each one.
- We document where each NIST AI RMF or ISO/IEC 42001 requirement is covered, and where gaps remain that your organization must own.
- Your team gets a handover session and the operating procedures needed to keep the AI management records current after delivery.
How we support
- Ongoing support, upgrades and continuous testing are available as retainers after the initial scope is delivered.
- Retainers can keep AI inventories, risk records and monitoring evidence current as systems change and as NIST revises AI RMF 1.0.
- Before any external assessment, confirm with the auditor or certification body which evidence they expect, since they decide what satisfies the requirement.
