The NIST AI Risk Management Framework (AI RMF) is free, voluntary guidance from the U.S. National Institute of Standards and Technology for managing the risks of AI. It organizes the work into four functions, Govern, Map, Measure and Manage, and a small business can put a useful version of it in place in a few weeks.
What the NIST AI RMF is, and what it isn’t
NIST released version 1.0 of the AI Risk Management Framework in January 2023. It was written for any organization that designs, develops, deploys or uses AI systems, from large software companies to a 20-person firm whose staff use ChatGPT to draft emails.
A few things it is not:
- Not a law. No federal law requires a private company to adopt it.
- Not a certification. There is no “NIST AI RMF certified.” If you need a certificate, look at ISO/IEC 42001.
- Not a checklist. It describes outcomes, not specific tools or settings, so you scale it to your size and risk.
That flexibility is the point. A hospital system and a marketing agency face very different AI risks, and the framework lets each focus on what matters to them.
What “trustworthy AI” means in the framework
The AI RMF describes the characteristics of trustworthy AI. In plain English, an AI system you rely on should be:
- Valid and reliable: it does what you need, consistently.
- Safe: it doesn’t put people at risk.
- Secure and resilient: it resists attacks and misuse, and recovers when something goes wrong.
- Accountable and transparent: someone is responsible, and people know when and how AI is used.
- Explainable and interpretable: you can understand enough about how it reaches results to trust them.
- Privacy-enhanced: it protects personal and confidential information.
- Fair, with harmful bias managed: it doesn’t treat people unfairly.
For most small businesses, the characteristics that come up first are security, privacy and accountability: where does our data go when staff paste it into an AI tool, and who decided that was acceptable?
The four functions: Govern, Map, Measure, Manage
Govern
Govern sets up the culture, roles and policies for managing AI risk. It applies across everything else. For a small business that usually means naming an owner for AI decisions, writing an AI acceptable use policy, and deciding how new AI tools get approved.
Map
Map is about understanding context: which AI systems you use, for what purpose, with what data, and what could go wrong. The practical starting point is an inventory. Most companies are surprised by how many AI tools are already in use without approval, which is why shadow AI discovery is often the first step.
Measure
Measure covers assessing and tracking the risks you mapped. That can be as simple as testing an AI tool’s output on real examples before rolling it out, checking a vendor’s security documentation, or reviewing a sample of AI-assisted work each month.
Manage
Manage is acting on what you found: prioritizing the biggest risks, putting controls in place, monitoring them and responding when something goes wrong. Examples include blocking unapproved AI tools, restricting which data can be shared with approved ones, and adding AI incidents to your incident response plan.
The Generative AI Profile
In July 2024 NIST published a companion document, the Generative AI Profile (NIST AI 600-1). It applies the framework to generative AI and lists risks that are unique to it or made worse by it. Several are directly relevant to everyday business use:
| Risk named in the profile | What it looks like at work |
|---|---|
| Confabulation | The tool confidently invents facts, citations or numbers. |
| Data privacy | Staff paste client or personal data into a tool that keeps or trains on it. |
| Information security | AI-generated phishing, or attacks that manipulate the tool through its inputs. |
| Intellectual property | Output that reproduces protected content, or your own confidential material leaking out. |
| Value chain and component integration | Risk inherited from the AI vendors and plug-ins you connect. |
If your team mainly uses generative AI assistants, this profile is the most practical part of the NIST material to read.
A practical starting plan for a small business
You don’t need a committee or a consultant army. A focused version looks like this:
- Weeks 1 to 2: inventory. List every AI tool in use, including browser extensions and features inside tools you already pay for. Note what data each one touches.
- Weeks 2 to 3: decide and document. Name an owner, approve the tools you’ll keep, and publish a short AI acceptable use policy with clear rules on what data can and can’t go into AI tools.
- Weeks 3 to 4: control. Turn on the business or enterprise versions of approved tools with the right data settings, and block or restrict the rest.
- Ongoing: measure and review. Check new tools before they’re adopted, review usage quarterly, and add AI to your vendor reviews and incident response plan.
That’s a credible Govern and Map foundation, with the beginnings of Measure and Manage. An AI readiness assessment can shortcut the first two steps if you’d rather have someone do the inventory and gap analysis for you.
Why it’s worth doing now
Customers increasingly ask about AI in security questionnaires. Cyber insurers are starting to ask too. And some state AI laws reference the NIST AI RMF as a recognized approach to managing AI risk. Having a documented, NIST-aligned program means you can answer those questions with evidence rather than reassurance.