Is the NIST AI RMF mandatory?

No. It is voluntary guidance. It is often used to structure an AI risk programme, and increasingly referenced in procurement and contracts, but it does not carry legal obligation the way the EU AI Act does.

What are the four NIST AI RMF functions?

Govern, Map, Measure and Manage. Govern is cross-cutting; Map establishes context, Measure analyses and tracks risk, and Manage acts on it.

Does AI RMF cover autonomous agents?

It applies, but its published guidance is model-centric. Agent risk is dominated by authority and action — which tools an agent can call and what it can irreversibly do — which needs controls the framework text does not specify.

LearnStandards

The NIST AI Risk Management Framework, in practice

The NIST AI Risk Management Framework (AI RMF) is a voluntary US framework for identifying and managing risk across the AI lifecycle. It is organized into four functions — Govern, Map, Measure and Manage — and is deliberately not a checklist: it describes outcomes to achieve rather than controls to install, which is its main strength and the reason teams find it hard to operationalize.

Updated 2026-09-18

The four functions

Govern is cross-cutting; the other three describe a cycle that repeats as systems and context change.

Govern — accountability, roles, policy and culture. The function everything else depends on.
Map — establish context: what the system is for, who it affects, what could go wrong.
Measure — analyse and track the risks you mapped, with methods appropriate to each.
Manage — act on what you measured: prioritize, respond, monitor and recover.

Applying it to agents rather than models

The AI RMF was written with models in mind, and most published guidance reflects that: training data, evaluation, bias, explainability. An autonomous agent shifts the risk surface. The question is less "is the output accurate" and more "what did this software do, with whose authority, to which system".

Mapping an agent therefore means mapping its authority: which tools it can call, which data it can reach, which actions are irreversible, who it acts on behalf of, and what happens when it delegates to another agent. Measuring means instrumenting decisions, not only outputs. Managing means having a control that can actually stop an action.

Where teams get stuck

Two places, consistently. The first is Map: an inventory that lists applications rather than agents and their authority, which makes everything downstream approximate. The second is Measure: no telemetry at the decision point, so risk is assessed from documentation rather than from what the system actually did.

Both are inventory problems before they are framework problems. A function cannot be performed against systems nobody has enumerated.

How it relates to other frameworks

AI RMF is voluntary and US-originated. ISO/IEC 42001 provides a certifiable management system covering similar ground. The EU AI Act is law rather than guidance, with obligations by risk tier. Using AI RMF to structure the work does not discharge EU AI Act duties, and certifying to ISO/IEC 42001 does not by itself satisfy them either.

Common questions

Is the NIST AI RMF mandatory?: No. It is voluntary guidance. It is often used to structure an AI risk programme, and increasingly referenced in procurement and contracts, but it does not carry legal obligation the way the EU AI Act does.
What are the four NIST AI RMF functions?: Govern, Map, Measure and Manage. Govern is cross-cutting; Map establishes context, Measure analyses and tracks risk, and Manage acts on it.
Does AI RMF cover autonomous agents?: It applies, but its published guidance is model-centric. Agent risk is dominated by authority and action — which tools an agent can call and what it can irreversibly do — which needs controls the framework text does not specify.