The NIST AI Risk Management Framework (AI RMF 1.0, published as NIST AI 100-1 in January 2023) is a voluntary framework for identifying, measuring and managing risk across the AI lifecycle. It is organized around four functions — Govern, Map, Measure and Manage — and is written to be adapted rather than certified against. NIST later published a Generative AI Profile (NIST AI 600-1) applying the framework to generative systems.
It has become procurement's common vocabulary. Because it is voluntary, sector-neutral and not a certification, it is the thing a security questionnaire can ask about without asserting a legal requirement — which is precisely why it appears in questionnaires so often.
Govern is where most programs are thin. Map, Measure and Manage describe work that a capable team can do in a document. Govern asks who is accountable, what the policies are, and how they are enforced — and that answer is either a system or it is a claim. The framework does not say which tools to use, so the burden of proof falls on whatever the organization can actually show.
How it differs from the EU AI Act. The AI RMF is voluntary guidance a firm chooses to adopt; the EU AI Act is binding law with classification tiers and obligations attached. They are frequently confused in procurement because both produce documentation. Only one of them carries penalties.
How it differs from model risk guidance. Model risk guidance is sector-specific supervisory expectation for regulated institutions. The AI RMF applies to anyone, and covers the lifecycle rather than the model.
Turning the framework into evidence
The RMF is a structure for organizing a governance program, not a checklist that produces artifacts. Its four functions tell an institution what to be able to demonstrate; they do not generate the demonstration. That gap is where most programs stall — the mapping is complete, the evidence behind it is a spreadsheet maintained by hand.
Closing it means treating enforcement records as the source of the evidence rather than a parallel system. Every decision carries the policy version that produced it, and the mapping is assembled from those records on request. That is the mechanism behind the compliance retrospector and the examination readiness suite, and it depends on rules being expressed as policy as code with tamper-evident records underneath. The framework sits alongside statutory obligations rather than replacing them — the EU AI Act approach covers the enforceable half. The difference between documenting a framework and enforcing it is worked through on the Credo AI comparison.
Related terms
SR 11-7 and SR 26-2 · Policy as code · AI sovereignty · Runtime governance · ISO/IEC 42001 · Full glossary
Put this in the path of your own agents.
Policy enforced inline between your agents and every model and tool they reach, with a record bound to the human who owns it.
Request a Demo Read the docs