NIST AI RMF in Plain English: How Small Teams Can Apply It
In short
- The NIST AI Risk Management Framework (AI RMF 1.0) is a free, voluntary framework from the US National Institute of Standards and Technology for managing the risks of AI systems. NIST says a revised version is in progress (as of November 2026).
- Its core is four functions: Govern (who decides and how), Map (understand the context), Measure (test and track) and Manage (act on what you found).
- A small team doesn't need a compliance department to use it: an AI inventory, one owner per system, a short risk review before launch and a few recurring checks cover most of the intent.
This guide is for founders, product leads and the person who ends up "owning compliance" in a small company that builds or uses AI. You'll get a plain-English explanation of the NIST AI RMF, what each part actually asks of you, and a lightweight way to apply it in about a month without hiring a governance team.
What the NIST AI RMF is (and isn't)
NIST released the AI Risk Management Framework 1.0 on January 26, 2023, after a public process that involved more than 240 contributing organizations over 18 months. It is intended for voluntary use: it isn't a law, a certification or a checklist you pass. NIST describes it as flexible and meant to augment the risk practices you already have.
The framework comes with companion resources on NIST's Trustworthy and Responsible AI Resource Center, including:
- The AI RMF Playbook: suggested actions for each part of the framework. Like the framework itself, it's voluntary; take what fits.
- The Generative AI Profile (NIST AI 600-1): published July 26, 2024, it applies the framework to risks that are specific to, or made worse by, generative AI.
- Sector work in progress: in April 2026 NIST released a concept note for an AI RMF profile on trustworthy AI in critical infrastructure.
As of November 2026, NIST's framework page states that AI RMF 1.0 is being revised as part of the White House AI Action Plan and that a revised version is in progress. The structure described below is the one in the published 1.0 version. If you adopt it now, keep an eye on the revision, but don't wait for it: the basic practices are unlikely to go out of date.
What "trustworthy AI" means in the framework
The AI RMF lists seven characteristics of trustworthy AI systems. They're a useful vocabulary for any risk discussion:
| Characteristic | A plain question to ask |
|---|---|
| Valid and reliable | Does it actually work, consistently, for the people and data it will see? |
| Safe | Could it lead to harm to people, property or the environment? |
| Secure and resilient | Can it withstand attacks and misuse, and recover when something fails? |
| Accountable and transparent | Is it clear who is responsible, and can people find out how it's used? |
| Explainable and interpretable | Can we explain what it does and why it produced a given output? |
| Privacy-enhanced | Does it respect people's privacy and handle personal data properly? |
| Fair, with harmful biases managed | Could it treat some groups unfairly, and are we checking? |
Two details matter in practice. First, NIST calls "valid and reliable" a necessary condition of trustworthiness: a system that doesn't work reliably can't be made trustworthy by documentation. Second, accountability and transparency relate to all the others. The framework also says that managing risk usually involves tradeoffs between these characteristics (for example, more data can help accuracy but increase privacy risk), so the goal is a reasoned balance, not perfection.
The four functions in plain English
The heart of the framework is the "Core": four functions, each broken into categories and subcategories. NIST is explicit that the actions "do not constitute a checklist," and that the functions can be performed in any order across the AI lifecycle. Govern runs across everything; Map, Measure and Manage apply to specific AI systems.
| Function | The question it answers | Small-team version |
|---|---|---|
| Govern | Who is responsible, and what are our rules? | A one-page AI policy, a named owner for each AI system, and rules for third-party AI tools and data. |
| Map | What is this system for, who does it affect, and what could go wrong? | An AI inventory and a short context note per system: purpose, users, data, vendors, possible impacts. |
| Measure | How do we know it's working and staying within limits? | Test before launch and on a schedule: accuracy on your own examples, security tests, and a log of incidents and complaints. |
| Manage | What do we do about the risks we found? | Prioritize, decide (fix, accept, avoid or transfer), keep a response plan, and review third-party AI regularly. |
A few of the official categories are worth knowing by name, because they map directly onto small-team habits: GOVERN 2 (accountability structures and training), GOVERN 6 (policies for third-party software, data and supply chain), MAP 1 (context is established and understood), MAP 5 (impacts on individuals, groups and society are characterized), MEASURE 1 (appropriate methods and metrics), MEASURE 3 (tracking risks over time), MANAGE 3 (risks from third parties are managed) and MANAGE 4 (response and communication plans are documented and monitored).
Applying it in about 30 days
Here's a practical sequence for a team of roughly 5 to 50 people. It's our suggestion, not an official NIST method; the framework leaves the "how" to you.
Week 1: Govern basics
- Name one person accountable for AI risk overall, and one owner for each AI system or AI-powered feature.
- Write a one-page AI use policy: which AI tools staff may use, what data may never be pasted into them, who approves new AI features, and how to report a problem.
- Decide your risk tolerance in plain words. The framework doesn't set it for you; it says organizations should define reasonable risk tolerance where no established guideline exists.
Week 2: Map what you have
- Build an AI inventory. Include what you build (a support chatbot, a scoring model) and what you buy (AI features in your CRM, coding assistants, transcription tools).
- For each item, record its purpose, users, data used (especially personal data), vendor, owner and the worst realistic outcome if it fails or is misused.
- Rank systems. NIST suggests giving higher initial priority to systems that interact directly with people or use sensitive data.
A simple inventory row can look like this:
| System | Owner | Data | Impact if wrong | Priority |
|---|---|---|---|---|
| Customer support chatbot (LLM vendor API) | Head of Support | Customer names, order history | Wrong refund promises, data shown to the wrong customer | High |
| Internal meeting summarizer | Ops lead | Internal meetings | Confidential notes stored with a vendor | Medium |
Week 3: Measure the high-priority systems
- Create a small test set from real (anonymized) cases and check accuracy before and after every significant change. NIST's guidance is that AI systems should be tested before deployment and regularly while in operation.
- Run basic security testing. For LLM apps, use the OWASP Top 10 for LLM Applications as your test plan.
- Check for obvious unfair outcomes across the groups your system serves, and note what you couldn't test.
- Set up a place to log incidents, user complaints and odd outputs, so you can track risk over time.
Week 4: Manage and repeat
- For each significant risk, record a decision: fix it, accept it (with a named person signing off), avoid it (don't ship the feature) or transfer it (contract terms with a vendor).
- Write a short response plan: how to switch the AI feature off, who tells customers, and who handles personal data incidents.
- Put a review on the calendar: quarterly for high-priority systems, and whenever you change a model, vendor or data source.
Use profiles to show progress
The framework describes "profiles" that tailor it to a setting. The most useful idea for a small team is the pair of Current Profile (how you manage AI risk today) and Target Profile (where you want to be). Comparing them shows the gaps, and the gaps become your roadmap. A one-page table with the four functions, "today", "target" and "next step" is enough, and it's a clear way to answer a customer's security questionnaire about AI.
If you build with generative AI, skim the Generative AI Profile (NIST AI 600-1) for risks and actions to add to your Map and Measure steps.
Common mistakes to avoid
- Treating it as a checklist to tick. NIST says the actions are not a checklist. Focus on the outcomes that matter for your systems.
- Forgetting the AI you buy. Third-party AI is in scope (GOVERN 6, MAP 4, MANAGE 3). Most small companies use far more AI than they build.
- Trying to eliminate all risk. The framework notes this can be counterproductive. Prioritize by impact and document what you accept.
- Doing it once. Models, vendors and data change. A short recurring review beats a long one-time document.
A note for Indonesian readers
The NIST AI RMF is a US framework, but it works well alongside Indonesian rules. If your AI processes personal data, the Personal Data Protection Law (UU PDP, Law No. 27 of 2022) applies, and your Map step is a natural place to record that. The Ministry's Circular Letter No. 9 of 2023 on AI Ethics (SE Menkominfo 9/2023) lists ethical values such as transparency, accountability, security and personal data protection that line up closely with NIST's trustworthy characteristics, so one set of documentation can serve both.
Next steps
- This week: name an owner and start the AI inventory, including the AI tools you buy.
- This month: write the one-page policy, test your highest-priority system, and record risk decisions.
- Next quarter: write a Current and Target Profile, and check whether NIST has published the revised AI RMF.
For the security side of AI systems, read our guide to the OWASP Top 10 for LLM Applications, or see our security services.
Want help setting up a lightweight AI risk process for your team? Email Binjesoft and we can help you build an AI inventory, a one-page policy and a first risk review. See our services for how we work.
Sources: NIST: AI Risk Management Framework · NIST AIRC: AI RMF 1.0 · AI RMF Section 3: AI Risks and Trustworthiness · AI RMF Section 5: Core · AI RMF Section 6: Profiles · AI RMF Section 1: Framing Risk · NIST AI 600-1: Generative AI Profile · JDIH Komdigi: SE Menkominfo No. 9 of 2023