The NIST AI Risk Management Framework
The NIST AI RMF is one of the most widely used frameworks for managing AI risk. This module shows how to apply it in practice.
Learning objectives
- Describe the purpose and status of the NIST AI RMF
- List the seven characteristics of trustworthy AI
- Explain and apply the Govern, Map, Measure and Manage functions
- Use the Generative AI Profile's twelve risks
- Explain how profiles tailor the framework
3.1 What the AI RMF is
The NIST AI Risk Management Framework (AI RMF 1.0), published by the US National Institute of Standards and Technology in January 2023, is a voluntary, sector-agnostic framework for managing AI risks. It's widely used internationally because it's practical, free and compatible with other standards. NIST has signalled that the framework is being revised, so check for updates.
3.2 Characteristics of trustworthy AI
The AI RMF describes seven characteristics of trustworthy AI systems:
- Valid and reliable (the foundation for the others)
- Safe
- Secure and resilient
- Accountable and transparent
- Explainable and interpretable
- Privacy-enhanced
- Fair, with harmful bias managed
3.3 The four core functions
GOVERN
Cultivate a risk management culture: policies, accountability, roles, workforce diversity and training, stakeholder engagement, third-party risk. Govern runs across all the others.
MAP
Establish context: intended purpose, users and affected people, legal requirements, potential benefits and harms, and the AI system's categorisation.
MEASURE
Analyse, assess and track risks with quantitative and qualitative methods: testing, evaluation, verification and validation (TEVV), including for trustworthiness characteristics.
MANAGE
Prioritise and act on risks: mitigate, transfer, avoid or accept; plan responses and recovery; monitor and communicate.
Each function breaks down into categories and subcategories. NIST's companion Playbook suggests concrete actions for each.
Govern: the complaints director is the system owner; the AI policy requires risk review before launch.
Map: purpose is routing, not deciding outcomes; affected people include vulnerable customers; consumer duty applies; potential harm: urgent complaints misrouted.
Measure: test routing accuracy on 2,000 historical complaints, broken down by complaint type and customer vulnerability flags; measure how often urgent cases are missed.
Manage: keyword safety net for urgent terms; human review of low-confidence cases; weekly monitoring; rollback plan.
3.4 The Generative AI Profile (NIST AI 600-1)
In July 2024, NIST published a Generative AI Profile applying the AI RMF to generative AI. It identifies twelve risks that are unique to, or made worse by, generative AI:
| 1. CBRN information or capabilities | 7. Human–AI configuration (e.g. over-reliance) |
| 2. Confabulation (hallucination) | 8. Information integrity (mis- and disinformation) |
| 3. Dangerous, violent or hateful content | 9. Information security |
| 4. Data privacy | 10. Intellectual property |
| 5. Environmental impacts | 11. Obscene, degrading and/or abusive content |
| 6. Harmful bias or homogenisation | 12. Value chain and component integration |
The profile then suggests actions for each risk, mapped to the four functions — a useful checklist for any organisation adopting generative AI.
3.5 Profiles and tiers
Organisations create profiles: tailored versions of the framework for a sector, use case or technology (like the GenAI Profile). NIST has also released concept work on a profile for trustworthy AI in critical infrastructure (2026). Using the RMF is about fitting it to your context, not ticking every subcategory.
Key takeaways
- AI RMF 1.0 (January 2023) is voluntary, sector-agnostic and widely used; a revision is under way.
- Seven trustworthiness characteristics, with 'valid and reliable' as the base.
- Govern runs across Map, Measure and Manage.
- The GenAI Profile (NIST AI 600-1, July 2024) lists twelve generative AI risks.
- Tailor the framework through profiles.