What is the NIST AI Risk Management Framework?
The NIST AI Risk Management Framework is a voluntary United States framework for identifying and managing risk in AI systems. It is organised around functions that cover governance alongside the mapping of context and the measurement of risk, with each function describing outcomes rather than prescribing controls. Adoption is not certified, which is why it appears inside procurement questionnaires rather than as a certificate.
Also called: NIST AI RMF, AI RMF
Why a voluntary framework shapes contracts it does not govern
Nothing in the framework is mandatory, and that has made remarkably little difference to how often suppliers are asked about it. Buyers in the United States, particularly agencies at federal level, have adopted its vocabulary in their questionnaires, so a supplier who cannot speak it answers every question from first principles. The framework is also deliberately outcome-based, which means it cannot be satisfied by purchasing a tool. An organisation using it well ends up with a risk register for each AI system and a named owner against each entry.
- Procurement questionnaires borrow the framework's vocabulary whether or not it is required.
- Outcome-based wording prevents the framework from being satisfied by a purchase.
- It produces a risk register per system, which is the artefact auditors ask for.
- Its vocabulary maps onto other regimes, reducing duplicated governance work.
How the NIST AI Risk Management Framework works
- Each AI system is placed in its context of use and its affected population.
- Risks are identified against that context rather than against the technology.
- Measurement methods are chosen per risk, with the evidence they produce stated.
- Treatment is assigned to a named owner with a review date.
- Governance arrangements are documented, covering escalation above a stated threshold.
- The register is reviewed as systems and their contexts change.
How the NIST AI Risk Management Framework is evaluated
Use of the framework is judged on the register it produces rather than on the policy that cites it. Examine the following across deployed systems:
- proportion of deployed systems with a current risk entry
- presence of a named owner against each identified risk
- evidence produced by each declared measurement method
- age of the most recent review per system
- coverage of affected populations in each context statement
Limitations, risks and common failure modes
- The framework is cited in a policy document that no deployment references.
- Risk entries are written at launch and never revised as usage widens.
- Measurement is declared without naming the evidence it is meant to produce.
- Context of use is described for the pilot rather than the eventual deployment.