NIST AI RMF explained: the four Functions

The NIST AI Risk Management Framework (AI RMF) is fast becoming the reference point for how organisations talk about AI risk, and it is often misunderstood. It is not a certification, and it is not a list of tools to buy. It is a taxonomy of outcomes for identifying, assessing and managing the risks of AI systems across their whole lifecycle, from design to decommissioning. Version 1.0, published in January 2023 as NIST AI 100-1, is voluntary, rights-preserving and designed to fit any organisation, sector or AI use case.

Think of the AI RMF less as an exam and more as a shared vocabulary. It gives your engineers, your executives, your customers and your regulators one way to describe where your AI programme is strong, where it is thin, and where you are heading next.

The four Functions

The AI RMF organises everything under four Functions. Govern is cross-cutting and sits over the other three:

• Govern: the culture, policies, roles and accountability that steer AI risk management across the whole organisation. This is the Function that puts AI oversight on the leadership table.
• Map: establish the context and frame the risks of a given AI system — who it affects, what it is for, and what could go wrong.
• Measure: analyse, assess and track the identified risks with quantitative and qualitative methods.
• Manage: prioritise the risks and act on them, allocating resources to the ones that matter most.

The genius of the AI RMF is that Govern is not a stage you finish, it is the layer that makes the other three trustworthy. Skip it and Map, Measure and Manage become box-ticking.

The seven trustworthy-AI characteristics

Underneath the Functions sit 19 Categories and 72 Subcategory outcomes, and they all serve a single goal: trustworthy AI. NIST defines that with seven characteristics — valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy-enhanced, and fair with harmful bias managed. A trustworthy AI system is not one that maxes out any single characteristic, it is one that balances them for its context.

That is the whole trick of the AI RMF. You are never scored pass or fail. You frame the risks of each system honestly, measure them against these characteristics, and manage the gap.

Why it pairs so well with ISO 42001

Because the 72 Subcategories are outcomes rather than prescriptions, they crosswalk cleanly to ISO/IEC 42001:2023 and other standards. If you already run a 42001 AI management system, most of the evidence you collect can be pointed straight at the matching RMF Subcategories. In Compliance One the full AI RMF 1.0 outcome library ships ready to assess each AI system against, cross-mapped to your 42001 controls, and the risk assessment turns into an owned, prioritised plan. The framework stops being a giant spreadsheet and becomes a picture leadership can read.