Briefing
Reference 4 · Primer · About 6 minutes
Download the MP3 · Leonard, an AI voice, reads a script written from the sources below.
Written from public sources and checked on 10 October 2026. The framework is paraphrased, not reproduced. Check the current text before relying on it.
The standards and frameworks that describe how to run AI governance as a system, and the practical work each one calls for.
These are Agile Insights AI Strategy & Governance offerings. Some are delivered with partners.
Sign up for our podcast at agile-insights.com.au.
Leonard The National Institute of Standards and Technology's AI Risk Management Framework is the plain language risk vocabulary that most AI governance, including ISO standards and many boards, quietly borrows. By the end of this episode, you will know what it contains, what it does not do, and where it stands as at October 2026.
Leonard First, what it is. The framework is published by NIST, a United States government standards agency. Version 1.0 appeared in January 2023. It is voluntary and sector neutral, meaning it suits a bank, a hospital or a retailer equally. It is not a law, and it is not certifiable. Nobody can audit you and hand you a certificate that says NIST compliant.
Leonard The problem it solves is a shared language. Picture a board asking whether a new credit model is safe. The data team talks about accuracy, legal talks about bias, security talks about attacks, and nobody answers the question. The framework gives everyone one set of words for risk, and one way to say who is accountable for each part.
Leonard Who uses it? Chief risk officers, chief information officers, privacy and compliance leads, and increasingly boards. It comes up when a customer sends a procurement questionnaire, when an insurer asks how you govern AI, when a regulator asks what you based your approach on, and when you write a board paper. Organisations with United States links often meet it first.
Leonard Now, what is in it. Four ideas carry the whole thing. The first is the four functions, which are Govern, Map, Measure and Manage. Govern is about culture, policy and accountability, and it wraps around the other three. Map is about understanding context: what the system is for, who it affects, and what could go wrong. Measure is about testing and tracking those risks with real evidence. Manage is about deciding what to do: treat, accept, or switch off.
Leonard The second idea is a definition of trustworthy AI. The framework names seven characteristics. A system should be valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy enhanced, and fair, with harmful bias managed. Valid and reliable is the foundation. The framework is candid that these pull against each other. More explainability can cost accuracy, so someone senior must choose the balance deliberately.
Leonard The third idea is that the framework is a family, not one document. The Playbook gives suggested actions for each outcome, and it is free on NIST's AI Resource Center. Profiles tailor the framework to a use case. The most useful today is the Generative AI Profile, published in July 2024 as NIST AI 600 dash 1. It lists twelve risks that are new or made worse by generative AI, including confabulation, which is made up output, plus privacy, information integrity and intellectual property.
Leonard The fourth idea is that it works at the level of outcomes, not procedures. It tells you what good looks like, and leaves the how to you. That makes it adaptable. It also means the phrase "aligned to NIST" can mean almost nothing unless someone shows evidence.
Leonard How does it connect to the rest of the stack? Think of three layers. Data foundations sit at the bottom. AI governance is in the middle, and this is where NIST and ISO slash I E C 42001 live. Local law and regulation sit on top. The two governance documents fit together well. ISO slash I E C 42001 is the management system you can be certified against. The NIST framework supplies the risk thinking that fills it. Roughly, Govern lines up with leadership and planning, Map with context and impact assessment, Measure with performance evaluation, and Manage with operation and improvement.
Leonard Getting started takes three steps. One, run a gap analysis of your current AI practice against the four functions, and be honest about the gaps. Two, build a register of your AI systems, each with a named owner, a purpose, and a risk rating. Three, take the Generative AI Profile risks and test them against the tools your staff are already using. Primary sources are the framework document NIST AI 100 dash 1, the Playbook, and the Resource Center, all linked in the show notes.
Leonard Now the currency check, as at October 2026. Version 1.0 is still the current text. NIST's own page says the framework is being revised, in connection with the White House AI Action Plan of July 2025, which asked NIST to remove references to misinformation, diversity, equity and inclusion, and climate change. NIST has given no timeline, and no new version has been published. Alongside, NIST has released a concept note, in April 2026, for a critical infrastructure profile. A Cyber AI Profile, linked to NIST's cybersecurity framework, has been out in draft form since December 2025. Agent security work is also under way, with an agent interoperability profile expected late in 2026, though I could not confirm its release. The core has stayed voluntary throughout. Check the current NIST pages before you cite it.
Leonard To recap: the NIST AI Risk Management Framework gives your organisation a common language for AI risk, with four functions and seven marks of trustworthy AI, and it pairs naturally with ISO slash I E C 42001. If you'd like to know more, sign up for our podcast at agile-insights.com.au.
All of the reference series (16) · Follow AI Governance Reference in Apple Podcasts or any podcast app
Everything from Agile Insights: briefings, Making You Think and the reference series