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 free, voluntary framework for managing AI risks. It is organised in four functions: Govern, Map, Measure and Manage. In July 2024 NIST added a Generative AI Profile (NIST AI 600-1) with risks specific to tools like ChatGPT and Claude. Use this checklist to see where your organisation stands.
Govern: culture, roles and policies
- ☐ A named owner is accountable for AI risk, with authority to approve or stop tools.
- ☐ An AI policy or acceptable use policy is published and staff have acknowledged it.
- ☐ Legal and regulatory requirements that apply to your AI use are listed (privacy law, sector rules, client contracts).
- ☐ Roles are clear: who requests a tool, who assesses it, who approves it.
- ☐ Staff receive AI training at least once a year.
- ☐ Third-party AI vendors are assessed before use and reviewed regularly.
- ☐ There is a way for staff to report AI problems without blame.
Map: context and risks of each use
- ☐ Every AI tool in use is recorded with its purpose, users and data types.
- ☐ Each use case states who could be affected (clients, employees, the public).
- ☐ Limits of the tool are documented (it can be wrong, biased or out of date).
- ☐ High-impact uses are flagged: decisions about people, legal or medical advice, financial reporting.
- ☐ Data flows are known: what goes in, where it is stored, whether it trains the vendor's model.
Measure: test and track
- ☐ Outputs are spot-checked for accuracy before important use.
- ☐ Bias and fairness are tested where AI affects people (hiring, lending, access to services).
- ☐ Security risks such as prompt injection and data leakage are considered, especially for AI agents.
- ☐ Metrics are agreed for important uses (error rate, time saved, complaints).
- ☐ Results of tests are recorded in the risk register.
Manage: act on the risks
- ☐ Each risk has a decision: accept, reduce, transfer or avoid.
- ☐ Human review is required before AI output reaches clients or is used in decisions about people.
- ☐ There is an incident process: who to tell, how to contain it, how to inform clients.
- ☐ Tools can be switched off or replaced quickly if a vendor changes terms.
- ☐ Risks are reviewed at least quarterly and after incidents.
Generative AI profile: extra checks
- ☐ Confabulation (made-up facts and citations) is addressed with review steps.
- ☐ Confidential and personal data are kept out of tools that train on inputs.
- ☐ Intellectual property and copyright of outputs are understood for each tool.
- ☐ Harmful or misleading content is filtered and AI-generated content is labelled where needed.
- ☐ Information security of the vendor is checked (see the vendor questionnaire).
Frequently asked questions
Is the NIST AI RMF mandatory?
No. It is voluntary guidance. Many companies use it because it is free, practical and recognised by clients and regulators.
How is NIST AI RMF different from ISO 42001?
NIST AI RMF is guidance on managing AI risk. ISO/IEC 42001 is a certifiable management system standard with formal requirements and audits. Many organisations use NIST to structure the work and ISO 42001 to certify it.