NIST AI RMF and ISO/IEC 42001: Which One Can Actually Be Certified

Finance functions building AI governance tend to reach for one of two reference points: the NIST AI Risk Management Framework, or ISO/IEC 42001. They are different kinds of thing, only one of them is certifiable, and neither is required by any law. Getting that straight first saves a lot of wasted effort.

What the NIST framework is

The document is NIST AI 100-1, titled Artificial Intelligence Risk Management Framework (AI RMF 1.0). It carries January 2023 on its cover and was released on 26 January 2023.

Its status is not in doubt, because NIST states it in the document itself: the framework is intended to be voluntary, rights-preserving, non-sector-specific and use-case agnostic. NIST's own announcement describes it as a guidance document for voluntary use by organizations designing, developing, deploying or using AI systems, and says it can help organisations in any sector and of any size.

So it is not law, not a standard in the formal sense, and not certifiable. There is no NIST certification and no NIST audit. The only route by which a finance team is ever genuinely required to use it is contractual: a customer, a parent company or a group policy requires it.

The four functions

The framework's core describes four functions. NIST's own descriptions are worth using rather than paraphrasing.

Govern cultivates and implements a culture of risk management within organisations that are designing, developing, deploying, evaluating or acquiring AI systems. Map establishes the context to frame risks related to an AI system. Measure employs quantitative, qualitative or mixed-method tools, techniques and methodologies to analyse, assess, benchmark and monitor AI risk and related impacts. Manage entails allocating risk resources to mapped and measured risks on a regular basis, as defined by the govern function.

One teaching point that gets lost when the four are presented as a sequence: govern is cross-cutting. It informs the other three rather than sitting first in a line. A governance structure that treats mapping, measuring and managing as stages to be completed in order has misread the framework.

Note also the word acquiring in the govern description. The framework is explicitly written to cover organisations that buy AI rather than build it, which is the position almost every finance function is in.

The companion documents, and what is not among them

NIST publishes several companions. The Generative Artificial Intelligence Profile is NIST AI 600-1, published 26 July 2024, described by NIST as a cross-sectoral profile of and companion resource for the AI RMF. The AI RMF Playbook, which suggests ways to navigate and use the framework, was completed on 30 March 2023. A Roadmap, a set of Crosswalks to other frameworks, and explanatory material were all published alongside the framework itself in January 2023.

Two corrections are worth making, because both errors are common.

There is no AI RMF 2.0. NIST's own resources page lists nothing published in 2025 or 2026, and the most recent companion remains AI 600-1 from July 2024. If a source refers to a 2025 or 2026 version of the framework, treat it as unverified. The page's own last-updated date refers to the webpage, not to a publication.

And the Generative AI Profile is a profile, not an update. It does not supersede or amend AI 100-1. Both remain current side by side, which means a reference to "the latest NIST AI guidance" is ambiguous and should be replaced with a document identifier.

One honest gap. AI 600-1 was issued under a United States executive order, and that order has been subject to later executive action. We have not verified its current status and make no claim about it. The NIST document itself remains published; the status of its originating order is a separate question.

ISO/IEC 42001 is a different proposition

Where the NIST framework is voluntary guidance, ISO/IEC 42001 is a management system standard, and it is certifiable through accredited certification bodies.

The strongest evidence of that is not a marketing page but another standard. ISO/IEC 42006:2025, first edition, published July 2025, is titled Requirements for bodies providing audit and certification of artificial intelligence management systems. It sets out additional requirements for bodies that audit and certify, extending ISO/IEC 17021-1 with AI-specific rules so that certification bodies assessing AI management systems against ISO/IEC 42001 operate with suitable competence and rigour, and it applies to the accreditation bodies assessing those certifiers.

ISO/IEC 17021-1 is the generic standard underpinning ISO 9001 and ISO/IEC 27001 certification. That ISO has extended it for AI management systems is hard evidence that ISO/IEC 42001 sits inside the standard management-system certification architecture rather than beside it, and that the accreditation infrastructure has existed since July 2025.

ISO describes ISO/IEC 42001 as the world's first standard for AI management systems, providing requirements and guidance for establishing, implementing, maintaining and continually improving an AI management system, covering leadership, planning, support, operation, performance evaluation and continual improvement across the AI lifecycle. Readers familiar with ISO 9001 or 27001 will recognise that clause sequence, which is why a finance or internal audit function already running one of those systems can extend existing governance rather than build something new.

We are deliberately not stating ISO/IEC 42001's publication date or edition here. ISO's own page for the standard would not serve on the attempts we made, and we are not willing to repeat a date we could not read at source. If you need to cite it formally, take the details from the standard's own front matter.

Three things certification is not

Certification is of a management system, not of a model, a tool or an output. ISO/IEC 42006's title is explicit on the point. No organisation is certified in respect of a particular model, and no AI output is ISO certified.

Certification is not EU AI Act compliance. There is no harmonised-standard status available on the public pages, and certification is voluntary and contractual. It discharges no obligation under the AI Act, as our piece on AI in the finance function and the high-risk regime explains.

And the controls commonly circulated as "the 42001 controls" come from the purchased standard, not from any public ISO page. For a programme that cites sources, either buy the standard and cite it properly or restrict claims to what the public pages say, which is very little.

Which to use

If the requirement is a shared vocabulary for talking about AI risk internally, the NIST framework is the most widely recognised one, costs nothing, and publishes crosswalks to other frameworks. If the requirement is to satisfy a customer, a tender or a group mandate, ISO/IEC 42001 is the one that can be certified and audited.

Neither is a legal obligation. The obligations that do apply to a finance function are elsewhere: the transparency and AI literacy duties under the EU AI Act, which are in force now and which we cover in what Article 4 actually requires, and the professional ethics duties that apply to accountants whatever tool they use. The framework document is published in full at NIST AI 100-1, and the discipline of separating a mandatory requirement from a recommended consideration is the same one our analysis of the IIA Standards applies to internal audit.