ISO 42001 vs NIST AI RMF: when a mid-market firm needs a certificate, not just a framework

This page is a buying comparison. It is for operations, quality, IT, compliance, and AI leads at mid-market manufacturers and professional-services firms who keep hearing two AI sentences in the same quarter. One customer, or an insurer, talks about the NIST AI Risk Management Framework (AI RMF). Another asks for an ISO/IEC 42001 certificate. Sales conversations will offer to "cover NIST and ISO" in one proposal. The useful question is a procurement question: which artifact does this cycle have to produce, and can you show the system behind it.

Those two artifacts are not the same product. The AI RMF is a voluntary framework: a shared language for AI risk, organized as Govern, Map, Measure, and Manage, plus profiles such as the Generative AI Profile. NIST's framework pages do not offer a certificate against it. ISO/IEC 42001 is a management-system standard. An accredited certification body can audit the artificial intelligence management system (AIMS) and issue a certificate for a defined scope. A profile can inform that system. It does not replace the certificate, and the certificate does not automatically fill a field that asked for the profile.

If the live fork is which certificate belongs on the RFQ, information security versus AI management is ISO 27001 vs ISO 42001, and quality versus AI management is ISO 9001 vs ISO 42001. This page does not invent prices, audit-day counts, accreditation identifiers, or a ranking of firms. Confirm the live schedule in the quarter you would book.

The decision here is narrower: an AI RMF profile this cycle, a 42001 certificate this cycle, or neither until the system exists. Then one honest request.

1. What each artifact actually is

Skip the clause lecture. Buyers do not award work because they memorized a function name.

The NIST AI RMF is guidance for managing risk to individuals, organizations, and society from artificial intelligence. NIST published AI RMF 1.0 as NIST AI 100-1 on 26 January 2023 (DOI 10.6028/NIST.AI.100-1). The NIST AI Risk Management Framework site is the public home for the document and the resources beside it. NIST describes the framework as intended for voluntary use, and as rights-preserving, non-sector-specific, and use-case agnostic. It is written for organizations that design, develop, deploy, or use AI systems. It does not prescribe a single control list.

The Core is four functions, then categories and subcategories. NIST's Core excerpt says the actions are not a checklist and not necessarily an ordered set of steps. Govern is the cross-cutting function: policies, accountability, culture, and the rules for how the other three run, including third-party software, data, and supply-chain issues. Map establishes context: purpose, intended use, the people affected, and impacts, including use beyond what you intended. NIST describes Map as the basis for an initial decision about whether to design, develop, or deploy. Measure uses quantitative, qualitative, or mixed methods to analyze, assess, benchmark, and monitor AI risk and related impacts, including testing before deployment and while the system is in operation. Manage allocates resources to the risks you mapped and measured, on the cadence Govern set. Risk treatment here includes plans to respond, recover, and communicate about incidents or events. The functions can be applied in the order that fits the lifecycle. After Govern, NIST says most users start with Map and then Measure or Manage, and that the work is iterative.

Part 1 of NIST AI 100-1 names characteristics of trustworthy AI: valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy-enhanced, and fair with harmful bias managed. Those are characteristics to discuss. They are not a score NIST issues.

A companion Playbook suggests ways to make the subcategories more actionable. NIST's Playbook FAQ says it is not a checklist, not an ordered list of steps, and not a document users are expected to implement in full.

NIST AI 600-1, the Generative AI Profile, was published on 26 July 2024 (DOI 10.6028/NIST.AI.600-1). NIST describes it as a cross-sectoral profile of, and companion to, AI RMF 1.0 for generative AI. It is there to help an organization identify risks that are particular to generative AI and to choose actions that fit its goals. It does not replace the Core. It is not a separate framework you "get" instead of the four functions, and it is not a certificate.

NIST's framework page also states that AI RMF 1.0 is being revised, and that on 7 April 2026 NIST released a concept note for a profile on trustworthy AI in critical infrastructure. Write the version you are actually using: AI RMF 1.0 (NIST AI 100-1), and NIST AI 600-1 when generative systems are in the boundary. A revision that is still in progress, and a concept note, are not documents a customer can score and not documents a certification body can audit.

ISO/IEC 42001:2023 is a management-system standard for artificial intelligence. ISO's catalogue entry describes requirements for establishing, implementing, maintaining, and continually improving an AI management system, for organizations developing, providing, or using AI. The same page separates that management-system standard from other AI documents that establish terminology, describe a machine-learning framework, or provide risk-management guidance. ISO's explainer says the standard does not replace laws or regulations. A certificate records that a certification body examined the management system. It does not promise that a model is accurate, fair, or lawful in every jurisdiction, and it does not approve a product.

ISO's certification page states that ISO develops standards and does not perform certification or issue certificates. Certification is an independent body giving written assurance that a system meets specified requirements. Accreditation is a further body recognizing that the certification body operates to international standards. For an AIMS, the standard your organization would be certified to is ISO/IEC 42001:2023. ISO/IEC 42006:2025 is a different document: additional requirements for the bodies that audit and certify an AIMS, supplementing ISO/IEC 17021-1. It is aimed at certification bodies and at accreditation bodies assessing them. You do not "get ISO/IEC 42006" as the plant's or the practice's certificate.

You can run an AIMS and also keep an AI RMF profile about the same systems. This page is about which one this request is for.

2. Who it suits

Read the field in the contract, the vendor portal, the insurance application, or the questionnaire. The logo on a slide is a clue. The sentence in the exhibit is the requirement.

NIST's own description is that use of the AI RMF is voluntary. A customer, an insurer, or your own board can still require it in a contract or in an internal mandate. If you are not under that sentence, the framework is a choice. Geography does not decide the artifact. A U.S. customer can require the certificate. A customer outside the United States can ask for the profile.

A customer or insurer mentions the NIST AI RMF when they want a common language: "aligned to the NIST AI RMF," "provide your AI risk profile," or "how do you govern, map, measure, and manage AI risk." They may also name generative AI, in which case the useful companion is NIST AI 600-1, not a new logo. They want scope, owners, gaps, and a version. They have not asked an accreditation body to stand behind a certificate.

A customer, portal, or insurer names ISO/IEC 42001 when they want a certificate they can trace, for a scope that matches the AI systems they rely on. A worksheet does not fill a field that says "ISO/IEC 42001 certificate." A 42001 certificate does not fill a field that says "AI RMF profile" or "NIST AI 600-1" unless that party has accepted the certificate as the evidence.

Put an AI RMF profile on this cycle when the written ask names the AI RMF, a profile, alignment, or the four functions, and does not name an accredited certificate on a date you can point at. Someone can walk which outcomes are in place, partial, or not started, for a named set of systems. If generative models draft engineering notes, quality procedures, client deliverables, or customer text, say so and use NIST AI 600-1 for that boundary. The 42001 ask, if any, is a forecast: a logo you hope to win, not a date that kills a contract this cycle.

On a plant, that profile is often one vision cell, a scheduling or maintenance model, and the generative tools a process engineer actually uses. The rest of the factory can stay out, if you write the boundary. In a professional-services firm, the profile is the models inside delivery: analysis, recommendations, drafts, or code a client will receive. A chatbot someone tried once is a different object. An internal copilot that never leaves the firm is in scope only when you choose to describe it.

Put ISO/IEC 42001 on this cycle when a customer, portal, insurer, or bid names that certificate, or an accredited AI management certificate, on a date you can point at. You can list the systems you develop, provide, or use, including third-party and generative models when they do the work you will claim. Someone owns policy, impact and AI risk, approval, monitoring, and retirement, and can be interviewed. The certificate boundary matches the systems the customer is buying. Pilots you will not claim stay out, with a reason.

The same plant and the same practice can be in either column. The split is the sentence in front of you, plus whether the system exists. A vision cell that accepts or rejects parts can sit in a quality process and still be the AIMS the customer named. A model inside a client memo can sit in how you deliver work and still be the AIMS. If the exhibit names the certificate, the profile will not clear it. If the exhibit names the profile, the certificate will not clear it unless they have said so in writing.

If neither description is true, build the system before you send. The person who writes the profile or the AI policy should not be the certification body.

3. Certifiable versus self-attested

The two words get used as if one product were "real" and the other were a draft. Both can be serious. They produce different evidence.

Self-attested, for this buy, means your organization describes its own alignment. You scope the systems, record a current picture against Govern, Map, Measure, and Manage, and add NIST AI 600-1 where generative systems are in that scope. You can do this with your own people. If you hire help, hire an implementer, and put their name on the file. NIST's framework pages do not offer a certificate, do not list accredited assessors for the AI RMF, and do not describe a conformity mark. A commercial review can still be a second set of eyes. Ask what they will hand you: a profile, a gap, or a letter. They cannot hand you a NIST certificate. If the letter says "certified to the NIST AI Risk Management Framework," slow down. There is no NIST directory of AI RMF certification bodies on the framework site. The verify guide is for the accreditation behind an ISO certificate.

Certifiable, for this buy, means an independent certification body can audit the AIMS and issue a certificate to ISO/IEC 42001:2023. ISO's explainer says that certification is voluntary, and that organizations choose it when they want independent confirmation. Voluntary in the standard is not the same as optional in your contract. Once a customer requires the certificate, the certificate is the buy. ISO still does not issue it. An accreditation body, in the United States often ANAB, assesses the certification body. IAF CertSearch is the database this site uses, together with the accreditation-body directory, to check certificates and accreditation. The method is the verify guide. A logo and a sister company's schedule are not that check.

A 42001 certificate is third-party. An AI RMF profile is not, unless a separate contract asks a named firm for a review and you are clear that the review is not a NIST certificate and not an ISO certificate. Do not let a proposal collapse those three into one line item.

4. How they map, and how they complement each other

The daily work can look similar, which is why the two products get treated as one purchase. Both expect a boundary, a way to decide risk, and evidence that people do what the description says: which systems and suppliers are in scope, who may approve a model, how you test and watch it, how you respond when it fails, and how you take it out of use.

The AI RMF describes outcomes. Subcategories sit under the four functions. The Playbook suggestions are not a required baseline. NIST AI 600-1 adds actions for generative AI. Those actions are still profile material. They become part of an audit only if your AIMS actually adopted them and the certification body is auditing ISO/IEC 42001, not "the NIST profile" as a standard.

The AIMS is what a certification body samples. Context, leadership, planning, support, operation, performance evaluation, and improvement are the management-system shape. For 42001, the content is AI: which systems are in scope, how you assess AI risk and impact, who may approve a system, how the lifecycle runs, and how you treat third-party and general-purpose models. Marking every control applicable, with no evidence, is a hollow document.

Use the functions as orientation, not as a finished crosswalk:

AI RMF function	What you are showing a customer	Where an AIMS usually has to be able to answer
Govern	Policy, roles, accountability, risk tolerance, third parties	Who owns the system, what the AI policy says, and which legal entity operates it
Map	Purpose, intended use, affected people, impacts	Which systems are in scope, and the impact records for the ones you will claim
Measure	Tests, metrics, trustworthy characteristics, tracking over time	How you monitor performance and risk after deployment, with records
Manage	Priority, response, recovery, communication, improvement	How you treat risk, what you do when a system fails, and how you retire it

That table is a buying aid. It is not clause-by-clause mapping, and it is not something to paste into an RFQ as audit criteria.

NIST's way to look up relationships is a crosswalk. The AI Resource Center crosswalk list says these are community submissions. Inclusion does not mean NIST endorses the other document, and it does not mean either document comprehensively covers the other. The list includes a crosswalk from the AI RMF to ISO/IEC 42001, with Microsoft as the provider. The file itself is titled as a map to ISO/IEC FDIS 42001, the final draft, not a statement that every row matches the published 2023 text. The same list has separate crosswalks for other AI documents, including ISO/IEC 23894, which is risk-management guidance. A guidance crosswalk is not a map onto a certificate. Use a crosswalk as a hint, then check each relationship against the control you operate and the evidence you could show.

A mapping can help one evidence set feed a profile and, later, an audit. It does not make a profile into a certificate, or a certificate into a current AI RMF profile. Do not paste a blog crosswalk into the request. Do not ask a certification body to invent a mapping you have not checked. Do not treat NIST AI 600-1 actions as controls a body must tick.

The boundaries have to tell the same story when both artifacts describe one operation. A profile scoped to "all AI in the company" and a 42001 scope scoped to "the vision cell on line two and the model that drafts the inspection procedure" are two systems. Write that split on purpose, or pick the boundary the customer is buying. Narrow on purpose is honest. Narrow by accident is how a customer later says the certificate misses the deliverable they asked about.

Where they complement each other, in practice:

• The profile is how you answer a customer who speaks Govern, Map, Measure, and Manage. The certificate is how you answer a customer who will not proceed without an accredited AIMS.
• Govern is the conversation that sounds most like leadership, roles, and policy. Map and Measure are the conversation that sounds most like impact, testing, and monitoring. Manage is the conversation that sounds most like treatment, incidents, and withdrawal. The AIMS still has to run as a system, not as four headings on a slide.
• NIST AI 600-1 belongs beside the Core when generative systems are in the boundary. It does not belong on the certificate as a second standard.
• One evidence binder is reasonable when both artifacts are real. Two outputs remain. Your organization owns the profile. The certification body signs only the certificate.
• A later certificate can reuse a profile that used the same legal entity, the same systems, and the same owner. A profile that describes a different boundary than the future certificate is work you will undo.

5. When a buyer needs a certification body

You need a certification body when the artifact this cycle is the certificate. You do not need one to write an AI RMF profile, to run a Playbook workshop, or to apply NIST AI 600-1. Hiring a certification body to "assess the NIST framework" produces a document NIST does not issue.

The certificate is the buy when a named party will not proceed without it: a portal, a contract, a bid, or an insurance application that names ISO/IEC 42001 or an accredited AI management system. The test is the sentence and the date. A competitor's homepage does not create the gate. A careful profile does not clear a field that says certificate.

Before you invite anyone, four facts have to be true. The scope is the systems the customer is buying, with pilots and personal tools out on purpose. A named owner can show the inventory, the impact records, and who may approve or retire a system. You will name the edition, ISO/IEC 42001:2023. You will verify that the legal entity's accreditation covers that edition. A 27001 schedule does not imply it. A 9001 schedule does not imply it. A brochure that says the group "does AI" does not imply it.

ISO/IEC 42006 tells you what to ask about the body, not what to print on your certificate. Ask which requirements the accreditation body used, then read the live schedule for ISO/IEC 42001. A training course on 42006 is not an accreditation. A product claim about a single model is not an AIMS certificate either.

Two profiles on this site already discuss ISO/IEC 42001 accreditation, for different buyers. The Schellman profile describes a public ANAB announcement for ISO/IEC 42001 and positions the firm on information security and AI management, not as a general factory registrar. The TÜV Rheinland of North America profile describes an ANAB scope that lists ISO/IEC 42001:2023 alongside quality, environmental, occupational health and safety, and information-security schemes. Read the profile that matches the work, then verify the legal entity this quarter. A profile is not the live schedule. This page does not rank either firm, and it does not claim that the schedule you need is still the one the profile described. The full set of directory profiles that state ISO/IEC 42001 on a published accreditation scope, including which mark is ANAB and which is UKAS only, is how to choose an ISO/IEC 42001 certification body in the US.

Four sequences cover almost every room. Pick one and write it in the first paragraph of the request.

Profile first, certificate later. The live ask is the framework. Finish a profile the owner can walk, including NIST AI 600-1 if generative systems are in scope. Mark 42001 as a later add, with a recheck date for live accredited scope. Do not ask a certification body to quote a certificate you will not staff.

Certificate first, profile later. The certificate is the gate. Certify the AIMS. A later profile can reuse the evidence and the boundary. Someone still has to write the outcomes in AI RMF language. The certificate does not generate that profile.

Both this cycle. Two requirements have dates, and both the AIMS and the profile match Tuesday. One evidence binder is fine. Two outputs remain. Decline a line that says the ISO audit certifies the AI RMF or NIST AI 600-1.

Neither this cycle. The requirement is real and the system is a slide. Four function names are not a profile. An AI policy nobody uses is not an AIMS.

6. How to scope the RFQ

One narrative, so the facts do not drift. Send a certification-body request only for the certificate. The AI RMF half is a document you prepare for the customer or insurer who asked for it. If only one artifact is this cycle, do not dress the other up as a line item so the proposal looks complete.

Fact	If ISO/IEC 42001 is this cycle	If the NIST AI RMF is this cycle
What you are buying	An accredited certificate for ISO/IEC 42001:2023	A profile against AI RMF 1.0 (NIST AI 100-1), plus NIST AI 600-1 if generative systems are in that boundary. Not a NIST certificate
Who may produce it	A certification-body legal entity accredited for that standard	Your team, or an implementer who will not also certify the AIMS. NIST does not certify AI RMF work on the framework pages
How you check them	The accreditation-body directory (ANAB is the usual U.S. mark) and IAF CertSearch, using the verify guide	There is no NIST AI RMF accreditation directory on the framework site. If a firm offers a certificate, ask which standard they are accredited to issue, then verify that claim
Boundary	AI systems you develop, provide, or use, the legal entity, the sites, and what stays out	The same systems, written as the profile scope. "All AI" is a scope choice you have to defend
What "done" looks like	Inventory, impact and AI risk records, a named owner, and a lifecycle you can walk	Outcomes an owner can walk under Govern, Map, Measure, and Manage, with the Generative AI Profile only where it applies
The other artifact	Later add, an input you already use, or not this cycle	Later add, or not this cycle
Where the customer looks	The certificate and the accreditation path	The profile itself, shared with that customer or insurer

Headcount is a band per location, not the about-page total. Name the same locations in the profile and in the AIMS scope when they are truly in both. For every certification body, ask which legal entity would issue the certificate, whether accredited scope for ISO/IEC 42001:2023 is live for your sites, how they restate the systems and the sites, and what a later add would require. Days, sites, and travel are what drives cost. This page does not publish a fee or a day count. Do not copy a duration from a blog onto the pack.

On the certificate request, you may say that customers also speak the AI RMF, so the body understands the language in your policies. Do not add "NIST AI RMF certification" or "NIST AI 600-1 certification" as a standard. Do not ask the body to assess a generative-AI profile as if it were a scheme. On the profile you hand a customer, name AI RMF 1.0, the date, the scope, the owner, whether NIST AI 600-1 is included, and whether a 42001 certificate already covers any of that scope.

7. Common mistakes, then one request

These are the misses that waste a quarter.

Treating 42001 as a badge that the model is acceptable. The certificate is the output of an audit of a management system. ISO says the standard does not replace the law. It does not make a vision model accurate, a client draft fit to send, or a use lawful. If counsel has not named the certificate, a headline about regulation is still a forecast.

Treating a profile, a workshop, or a "certified to NIST" letter as the certificate. The framework pages offer guidance and profiles. They do not offer the diploma the letter implies. Ask which accredited standard the firm can actually issue. If the answer is a workshop, you bought a workshop.

Pasting the FDIS crosswalk into the pack. The community file on the AI Resource Center maps to the final draft and carries NIST's non-endorsement limit. It is a hint for people who already know both documents. It is not the control decision for your system, and it is not the audit plan.

Putting the framework on the RFQ as if it were a scheme. "Quote ISO/IEC 42001 and NIST AI RMF certification" asks a body to sell a product NIST does not certify. Quote the certificate only when the AIMS exists. Keep the profile as your document.

Inferring accreditation. A 27001 line, a 9001 line, a group press release, or a 42006 course does not put ISO/IEC 42001 on the schedule of the legal entity that would sign your certificate. Check this quarter. The two directory profiles in section 5 are starting points for reading, not a substitute for that check.

Scoping "all AI." Every copilot, every pilot, and every personal account makes a boundary you cannot walk. Name the systems that do the work you will claim. Say what is out.

Hiring the author as the certifier. The firm that drafted the inventory and the policy should not be the firm that issues the certificate. Implementation and accredited certification are different jobs.

Skipping the version. "Aligned to NIST" with no number is not a profile. Name AI RMF 1.0, and name NIST AI 600-1 when generative systems are in scope. Do not cite a revision that is still in progress, or the April 2026 concept note, as if either were the requirement.

Swapping artifacts. A profile does not clear a certificate field. A certificate does not clear a profile field. Overlap is not acceptance.

Make the call in one sitting. The output is a sentence at the top of the request.

1. Write the actual requirement in their words. If one field says an AI RMF profile and another says an ISO/IEC 42001 certificate, note which deal is blocked this cycle.
2. Match it to a system you can show. A profile needs a scoped current picture an owner can walk, including NIST AI 600-1 if generative systems are in it. A certificate needs an AIMS: inventory, impact and AI risk records, and a named owner. If that system does not exist, do not send.
3. Name one primary artifact. Mark the other "later add" or "not this cycle," unless both are real enough to defend.
4. Name the boundary, the owner, and the sites. Say whether the profile and any certificate describe the same legal entity and the same systems. Separate the plant systems from the delivery systems if they are not the same claim.
5. Check the signer before you invite them. For 42001, use the verify guide. For a profile, confirm you are not buying a NIST certificate. This site's directory is an editorial shortlist of certification bodies for the ISO standards it tags. It is not a list of AI RMF assessors. A 42001 tag does not mean the organization assesses the framework.

If the room is still split after step 3, do the work that unblocks the deal and park the other until its evidence exists.

When the certificate is the artifact, send one request. The Get quotes form lists the ISO standards on the directory desk. The NIST AI RMF is not one of those checkboxes. In Scope notes, say whether a customer asked for an AI RMF 1.0 profile, state whether NIST AI 600-1 is included, and say whether ISO/IEC 42001 is this cycle or a later add. If the form will not send without a ticked standard, do not let a ticked ISO/IEC 42001 box stand in for a profile-only request. Put any certificate you already hold in Existing certificates. Submit once. The artifact on the page has to be the artifact you can staff.