NIST AI RMF 1.0.

What the AI RMF actually is, in plain English.

The NIST AI Risk Management Framework: NIST AI 100-1: was published on 26 January 2023 in response to the National AI Initiative Act of 2020. NIST built it the way NIST builds frameworks: through ~18 months of multi-stakeholder workshops, three RFI rounds, and consensus drafting. The result is a voluntary, non-sector-specific, non-prescriptive framework structured in two parts: Part 1: Foundational Information (risk framing, trustworthy-AI characteristics, AI lifecycle, AI actors) and Part 2: Core & Profiles (the four functions, 19 categories, 72 sub-categories, and the Profile concept).

The four core functions are the operational unit of the RMF: Govern (the cultivate-a-risk-aware-culture function, cross-cutting), Map (establish context and identify AI risks), Measure (analyze and assess), and Manage (allocate resources and respond). Govern sits at the center, intersecting all three operational functions. Each function decomposes into categories (e.g. Govern 1.1, Govern 1.2…), and each category into sub-categories: the granular outcomes you actually evidence.

The framework is supplemented by a Playbook (an interactive companion offering suggested actions, references, and documentation per sub-category), a series of Profiles (published or self-built: the Generative AI Profile NIST AI 600-1 is the most consequential), and the AI Resource Center (AIRC): NIST's hub for crosswalks (to ISO 42001, OECD AI Principles, EU AI Act, etc.), case studies, and ongoing updates.

What the RMF is not: a certifiable standard. There is no certification body. There is no accredited audit. There is no public registry of "RMF-conformant" organizations. NIST is explicit about this in the framework itself: the RMF is a resource, not a compliance regime. That distinction makes it powerful (you can adopt without procurement friction) and limited (a buyer asking for proof of trustworthy-AI maturity needs more).

Should you adopt RMF first?

Three questions. The answer tells you whether NIST AI RMF is a strong starting point, or whether you should jump directly to a certifiable standard.

Govern, Map, Measure, Manage.

The RMF Core is four functions that operate continuously and iteratively across the AI lifecycle. Govern is foundational and cross-cutting; the other three are operational and run in roughly that order: but in practice they loop.

The trustworthy-AI characteristics: valid & reliable, safe, secure & resilient, accountable & transparent, explainable & interpretable, privacy-enhanced, fair with harmful bias managed: are the substantive frame the framework's outcomes are measured against. Auditors operating outside the RMF (ISO 42001 CBs, EU AI Act notified bodies) borrow these directly.

A tour of the 72 sub-categories.

Click each function below for the categories an RMF-aligned program produces evidence against. The Playbook expands every sub-category with suggested actions, documentation, and references: including ISO 42001 Annex A crosswalks.

NIST AI 600-1: the generative-AI overlay.

On 26 July 2024, NIST published AI 600-1: the AI RMF Generative AI Profile: in response to the Biden Executive Order 14110. It is the most operationally useful artifact NIST has published on generative AI to date. The Profile identifies twelve risks unique or amplified by GenAI and maps them to specific RMF Core sub-categories with concrete suggested actions.

The twelve GenAI-specific risks are: CBRN information, confabulation (hallucination), dangerous, violent, or hateful content, data privacy, environmental impacts, harmful bias / homogenization, human-AI configuration, information integrity, information security, intellectual property, obscene, degrading, abusive content, and value chain & component integration. For each, the Profile lists suggested actions across all four core functions plus references to additional NIST work (red-teaming guidance, eval methods, transparency artifacts).

The Profile is doing real work in 2026. Federal agencies under OMB M-24-10 reference it directly. Enterprise procurement teams cite it in security questionnaires. ISO 42001 CBs use it as the de-facto baseline for what "responsible GenAI" looks like under Annex A.5 and A.6. If you ship GenAI features, your team should know the twelve risks by name.

Three ways organizations use the framework.

RMF is non-prescriptive by design: you tailor it to your context. The three usage patterns we see, in order of program maturity:

From kickoff to a defensible RMF Profile: 10 – 16 weeks.

RMF on its own is fast: no audit, no CB. The work is substantive: identifying AI systems, framing risks, picking outcomes, building evidence, training the org. We sequence it so a 42001 readiness can pick up directly from the Profile output.

RMF is the methodology. We make it operational.

Every Nexurion RMF engagement is led by a senior practitioner: the person on the engagement letter is the person walking your AI inventory, sitting in your risk-committee reviews, and on the call when leadership needs to make a tradeoff between shipping speed and risk acceptance. Read our methodology »

We use the NIST Playbook as a starting point for every sub-category, then tailor: not every suggested action fits every org, and a Profile copy-pasted from the Playbook is exactly the wallpaper we don't ship. Every outcome we accept gets a defensible rationale; every outcome we defer gets a roadmap entry with an owner and a date.

The artifacts that come out of an RMF engagement are reusable: AI policy, AI inventory, impact assessments, risk register, trustworthy-AI scorecards, third-party-risk DD records, monitoring KPIs, and the Profile document itself. Each one is built once and feeds 42001, EU AI Act conformity assessments, OMB M-24-10 evidence, and enterprise security questionnaires.

Six places an RMF program becomes wallpaper.

RMF's voluntary, non-prescriptive design is its great strength and its great weakness. Without discipline, alignment becomes self-deception.

RMF against the rest of the AI stack.

NIST publishes crosswalks for the major frameworks; we maintain our own with finer granularity. The shape that matters: RMF tells you the substance, the others tell you the shell.

RMF in year four.

The RMF was published in early 2023. The GenAI Profile arrived July 2024. Three things to watch in 2026:

• Sector-specific Profiles. Healthcare, financial services, and HR communities are publishing tailored Profiles. Where one exists for your sector, start there.
• Federal alignment. OMB M-24-10 obligations on agencies cascade to vendors via FAR / contract terms; RMF alignment is becoming a procurement floor for federal AI work.
• Updated GenAI guidance. NIST has signaled additional GenAI evaluation guidance in the AI RMF Knowledge Base: particularly around evaluation methodology, red-teaming, and content provenance.

Read our deeper take in Field Notes Vol. III.

RMF rarely stands alone.

In order of how often the question comes up alongside it.

Frequently asked.

Field notes on NIST AI RMF.

Pieces from The Field Notes directly relevant to the framework.