Is the NIST AI RMF mandatory for US companies?

No. NIST publishes the AI RMF for voluntary use, and there is no NIST certification against it. It can still matter commercially: customers, partners, and auditors may ask how you manage AI risk, and the framework gives you a recognized structure to answer with. Check with counsel whether any federal, state, or sector rules apply to your specific AI uses.

What is the difference between the AI RMF and NIST AI 600-1?

AI RMF 1.0 (NIST AI 100-1) is the general framework for any AI system, built around Govern, Map, Measure, and Manage. NIST AI 600-1, the Generative AI Profile released in July 2024, applies that framework to generative AI. It names 12 risks such as confabulation and information security and lists suggested actions under each AI RMF subcategory.

Where should a mid-size company start with the AI RMF?

Start with Govern and Map, not tools. Name an accountable executive and a small working group, build an inventory of every AI system including vendor features, and sort uses into risk tiers. Then focus testing, documentation, and shutdown planning on the high-tier systems that affect customers, employees, or decisions about people.

Do we have to complete every subcategory in the framework?

No. NIST says its companion Playbook is neither a checklist nor a set of steps to follow in its entirety, and the framework expects effort to match your risk tolerance and priorities. Use the subcategories as a menu: apply them fully to high-risk systems, lightly to low-risk ones, and record why you made those choices.

NIST AI RMF in Plain English: Govern, Map, Measure, Manage

Q: Is NIST updating the AI Risk Management Framework?

Yes. NIST's AI RMF page says version 1.0 is being revised as part of the White House AI Action Plan, which directs NIST to remove references to misinformation, diversity, equity and inclusion, and climate change. Until a revised version is published, 1.0 remains the current text, so organize your evidence by function and subcategory ID to make updating easier.

The NIST AI Risk Management Framework (AI RMF 1.0, published as NIST AI 100-1) is free, voluntary guidance for managing the risks of AI systems. It organizes the work into four functions: Govern, Map, Measure, and Manage. Its Generative AI Profile (NIST AI 600-1) adds 12 risks specific to generative AI. NIST says a revision is underway, and version 1.0 remains the published text.

At a glance

Document	What it is	Released
AI RMF 1.0 (NIST AI 100-1)	The framework itself	January 26, 2023
AI RMF Playbook	Suggested actions per subcategory	Companion to 1.0
Generative AI Profile (NIST AI 600-1)	12 GenAI risks plus actions	July 26, 2024
Cyber AI Profile (NIST IR 8596)	Cybersecurity Framework profile for AI	Preliminary draft, December 16, 2025

What the AI RMF is, and what it is not

The AI RMF is guidance, not regulation. NIST describes it as intended for voluntary use, and there is no NIST certification to earn against it. Its value is a shared structure and vocabulary that legal, security, data, and business teams can all work from, and a credible way to show customers, auditors, and boards that AI risk is handled on purpose rather than by luck.

It is written for organizations that design, develop, deploy, or use AI. That last word matters for a mid-size company: even if you only switch on AI features inside software you buy, you are one of the "AI actors" the framework addresses.

The framework defines seven characteristics of trustworthy AI:

Valid and reliable
Safe
Secure and resilient
Accountable and transparent
Explainable and interpretable
Privacy-enhanced
Fair, with harmful bias managed

NIST is explicit that these involve tradeoffs, and that rarely do all characteristics apply in every setting. Part of the job is deciding which matter most for each system.

This article is general information, not legal advice. If AI in your business touches hiring, lending, housing, insurance, or health decisions, ask counsel which federal and state laws apply.

The four functions in plain English

The framework's Core has four functions, each broken into categories and then subcategories with IDs such as GOVERN 1.6. Use those IDs to file your evidence, because they give auditors and partners a common reference.

Govern: decide who owns AI risk, and how

Govern is the cross-cutting function; NIST says it is infused throughout the other three. Its six categories cover policies, accountability, workforce diversity and culture, engagement with people outside the build team, and third-party risk. Plain-English questions:

Who can approve a new AI use, and who can stop one?
Do we keep an inventory of AI systems? (GOVERN 1.6)
How do we retire an AI system safely? (GOVERN 1.7)
What do our policies say about vendor and third-party AI? (GOVERN 6.1)

Map: understand the system in its context

Map "establishes the context to frame risks related to an AI system." Its five categories cover purpose and setting, categorization of the system, capabilities and expected benefits and costs, third-party components, and impacts on people and organizations. Ask: What is this system for, and what is it not for? Who is affected if it is wrong? What do errors cost, in money and otherwise (MAP 3.2)? Where does a human review the output (MAP 3.5)?

Measure: test it, then keep testing

Measure uses quantitative, qualitative, or mixed methods to assess the risks you mapped, across four categories. MEASURE 2 is the heart of it: evaluating each trustworthiness characteristic, from validity and reliability (MEASURE 2.5) and security and resilience (MEASURE 2.7) to privacy (MEASURE 2.10) and fairness and bias (MEASURE 2.11). MEASURE 3 tracks risks over time, and MEASURE 4 checks whether your measurements themselves work.

Manage: act on what you found

Manage "entails allocating risk resources to mapped and measured risks on a regular basis." Its four categories cover prioritizing and treating risks, maximizing benefits, managing third-party risk, and monitoring and communication after deployment. Two subcategories that mid-size firms often skip: MANAGE 2.4, which calls for clear responsibility and mechanisms to "supersede, disengage, or deactivate" a system that behaves outside its intended use, and MANAGE 4.3, which asks that incidents and errors are communicated to the people affected.

NIST's companion Playbook lists suggested actions for every subcategory, and it says plainly that the Playbook "is neither a checklist nor set of steps to be followed in its entirety." Pick what fits your risk.

The Generative AI Profile in one table

NIST AI 600-1 is a cross-sector profile for generative AI. It names 12 risks that generative AI creates or makes worse, then lists suggested actions under each AI RMF subcategory, with action IDs such as GV-1.1-001.

Risk	In plain English
CBRN information or capabilities	Easier access to weapons-related knowledge
Confabulation	Confident but false output, often called hallucination
Dangerous, violent, or hateful content	Easier creation of harmful content
Data privacy	Leakage or misuse of personal data
Environmental impacts	Energy and resource use of training and running models
Harmful bias or homogenization	Amplified bias and overly uniform outputs
Human-AI configuration	Over-reliance, automation bias, misplaced trust
Information integrity	Content that blurs fact and fiction at scale
Information security	New attack paths such as prompt injection and data poisoning
Intellectual property	Reproducing protected material, exposing trade secrets
Obscene, degrading, and/or abusive content	Synthetic abusive imagery
Value chain and component integration	Poorly vetted third-party models, data, and tools

For a typical company rolling out chat assistants and copilots, the risks that come up first are usually confabulation, data privacy, information security, human-AI configuration, and the value chain.

A starter playbook for a mid-size US company

You do not need a dedicated program office to begin. The sequence below suits a company with a mix of bought and built AI, and each step maps to the framework so your evidence lines up later.

Step	AI RMF anchor	Evidence to keep
1. Name an owner and a working group	GOVERN 2.1, GOVERN 2.3	Charter and named roles
2. Build an AI inventory	GOVERN 1.6	Systems, owners, vendors
3. Sort uses into risk tiers	GOVERN 1.3, MAP 5.1	Tier and reasons per system
4. Publish an AI use policy	GOVERN 1.2, GOVERN 2.2	Policy and training records
5. Map each high-tier system	MAP 1 to MAP 4	One-page system card
6. Test before go-live	MEASURE 2	Test sets, results, sign-off
7. Plan response and shutdown	MANAGE 2.4, MANAGE 4.1	Runbook and named owner
8. Review on a schedule	GOVERN 1.5, MEASURE 3	Review notes and changes

Steps 1 and 2: owner and inventory. Put one senior leader on the hook (GOVERN 2.3 places responsibility for AI risk decisions with executive leadership) and form a small group from IT, security, legal, HR, and the business. Then list every AI system: chat assistants, copilots, AI features inside SaaS tools such as your CRM and HR platform, internal models, and agents. Vendor AI counts.

Step 3: tiers. Keep it simple. Low: drafting and summarizing internal content, with a human editing everything. Medium: customer-facing content, or internal decisions where errors are visible and reversible. High: anything that affects people's access to jobs, credit, housing, insurance, health care, or legal outcomes, plus any system that takes actions on its own. Effort follows the tier, which is what GOVERN 1.3 asks for.

Step 4: policy. A short AI use policy covers approved tools, data that must never be entered, human review rules, and how to report problems. Our guide on what to include in an AI use policy gives a template structure.

Step 5: map. For each high-tier system, write a one-page card: purpose, out-of-scope uses, users and affected people, data sources, vendor and model, known failure modes, and where a human checks the output.

Step 6: measure. Test on your own data before launch: accuracy on a sample you have labeled, a confabulation check using questions with known answers, outcome comparisons across groups where people are affected, and adversarial tests such as prompt injection for anything connected to email, documents, or tools. Record the results and who accepted them.

Step 7: manage. Decide to deploy, limit, or stop. Name who can switch the system off and how, write an incident runbook, and agree how affected customers or staff will be told.

Step 8: review. Re-test after model or vendor changes and on a fixed calendar, and keep short notes of what changed.

Where the framework is heading

Two developments matter for planning.

Revision. NIST's AI RMF page states that the AI RMF 1.0 is being revised as part of the White House AI Action Plan. The July 2025 America's AI Action Plan directs NIST to revise the framework to remove references to misinformation, diversity, equity, and inclusion, and climate change. When this article was checked, NIST's pages still described the revision as in progress. If your program is organized by function and subcategory ID, an update means reviewing your mappings, not starting again.

Related profiles. In December 2025 NIST released a preliminary draft Cyber AI Profile (NIST IR 8596), which applies the Cybersecurity Framework to AI. In April 2026 it published a concept note for an AI RMF profile on trustworthy AI in critical infrastructure. Security teams already using the Cybersecurity Framework should follow the Cyber AI Profile as it matures. NIST also publishes crosswalks that map the AI RMF to other frameworks and standards.

Mistakes to avoid

Starting with tools. Buying an evaluation platform before you have an inventory and owners puts Measure ahead of Govern and Map.
Ignoring vendor AI. Much of a mid-size company's AI arrives inside software it already pays for.
Treating every system the same. Tiering keeps effort proportionate.
Writing a policy nobody reads. Short, specific, trained, and enforced beats long and generic.
No off switch. If nobody knows who can disable a system, you have not met MANAGE 2.4.

How CertTulen Academy can help

CertTulen Academy, a Microsoft Training Services Partner, runs live, instructor-led AI training online. For managers who will own AI adoption and governance, the one-day AB-731 Drive AI Transformation in Your Organization course is aligned to Microsoft's AI Transformation Leader certification. For a broader grounding in generative AI for business leaders, see the two-day Generative AI Leader course. Leadership teams can book a private cohort, or contact us for a batch that suits your time zone.