US governance / NIST AI RMF

Four functions.
One operating system.

A concrete mapping from Govern, Map, Measure and Manage to production AI architecture and records. Published 25 August 2026.

Use the framework as an operating model

NIST AI RMF 1.0 is voluntary and use-case agnostic. Its value is not a badge; it is a common language connecting executives, product, engineering, risk and affected stakeholders. Production implementation should attach each outcome to an owner, system behaviour, measurement and evidence source.

Map the functions to system layers

Govern becomes policy ownership, inventory, risk acceptance and change control. Map becomes the workflow, actors, impact, data and deployment boundary. Measure becomes evaluation sets, monitoring, intervention and independent review. Manage becomes release gates, incident response, rollback and prioritised treatment.

Connect GenAI risks

For generative systems, extend the record to prompt and context provenance, content risks, hallucination and abstention, privacy, misuse, model supply chain and human factors. The NIST Generative AI Profile provides a useful risk catalogue, but controls still need to be tailored to the actual task.

Primary sources

NIST AI RMF 1.0 · NIST AI 600-1.