To adopt the NIST AI RMF yourself, start with Govern to set policy and roles, then Map your AI systems and their context, Measure risks with tests and metrics, and Manage them with treatments, monitoring and incident response. The framework is voluntary and no one certifies you against it.
What are you adopting?
NIST published AI RMF 1.0 on 26 January 2023 to help organisations manage risks to people, organisations and society from AI. NIST says it is voluntary, is being revised, and has a Playbook of suggestions that is not a checklist. The framework describes trustworthy AI as valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy enhanced, and fair with harmful bias managed. A generative AI profile, NIST AI 600-1, adds suggested actions for that use.
Because there is no audit, "done" means you can show your process works. Aim for a living program rather than a project with an end date.
The plan
Step 1: sponsor and owner. Get an executive sponsor and name a program owner. GOVERN 2.1 asks for clear roles and accountability.
Step 2: discover your AI. Survey teams, check procurement and cloud bills, scan code for model SDKs. Build the inventory (GOVERN 1.6).
Step 3: write the AI policy. One or two pages. Cover the trustworthiness characteristics, acceptable use, approval for new systems, and who can pause one (GOVERN 1.2).
Step 4: set risk tolerance. Decide what you will accept, what needs executive approval and what you will not do. Record it (GOVERN 1.3 and 1.4).
Step 5: map each system. For each: purpose, users, context, data, third party components, legal constraints, and potential harms and benefits (MAP 1 to 5). Rank systems by risk so you spend effort where it counts.
Step 6: define measures and test. For the higher risk systems, choose metrics and thresholds and run tests for performance, resilience to bad input, security, privacy and bias where relevant (MEASURE 1 and 2). Save results.
Step 7: treat and monitor. For each documented risk, pick a treatment: mitigate, transfer, avoid or accept with sign off (MANAGE 1). Set up production monitoring and a deactivation path (MEASURE 2.4, MANAGE 2.4, MANAGE 4.1).
Step 8: handle third parties. Review model and data providers, contract terms and fallback plans (GOVERN 6.1, MANAGE 3.1).
Step 9: incidents and learning. Log incidents and errors, communicate them to the right people and feed lessons back into steps 4 to 7 (MANAGE 4.3).
Step 10: review. Review the inventory, risks and policy on a calendar, and when systems change or NIST updates the framework.
Who does what in a small team?
The executive sponsor approves policy and risk tolerance. The program owner runs the register and reviews. Engineering leads own tests, monitoring and the deactivation path. Legal or compliance reviews contracts and legal constraints. Product owners write the system profiles and impact notes.
What evidence will you have?
Policy with approval date, roles table, inventory exports, system profiles, risk register with treatments, test reports, monitoring dashboards, vendor reviews, incident log, review minutes.
What goes wrong in a first attempt?
Three things recur. The first is boiling the ocean: trying to assess every model and every tool at once. Rank systems by potential harm and begin with the top few. The second is a policy with no owner, which nobody follows. Give each system a named owner and a review date. The third is measuring only accuracy. NIST's trustworthiness list is wider, and a system can be accurate and still unsafe, unfair or leaky, so choose at least one measure from several characteristics for higher risk systems.
Keep the effort proportionate. A team using a hosted model for an internal summarising tool needs a lighter file than a team whose model influences decisions about people. The framework expects risk based choices, and your written reasoning is part of the evidence.
What to do
- Appoint the sponsor and owner.
- Build the AI inventory.
- Approve policy and risk tolerance.
- Map and rank systems.
- Test the highest risk ones first.
- Add monitoring and a switch off path.
- Review vendors and contracts.
- Log incidents and review on a schedule.
Quick check
Where should I start?
With Govern: policy, roles and an inventory. Map, Measure and Manage depend on knowing what AI you have and who owns it.
Can I get certified against the AI RMF?
NIST's page describes the framework as voluntary and lists no certification. Any third party review would use its own method.
How Safeguard helps automate this
The inventory and monitoring steps are where manual work piles up. Safeguard's compliance suite covers frameworks, controls, evidence, continuous monitoring, automated tests and questionnaires, so evidence is collected from connected systems and code instead of by screenshot, and monitoring runs continuously between reviews, saving the team time for other work. Technical evidence comes from your real code and dependencies, using scans, SBOMs and reachability, which helps with discovery of AI libraries and the third party component record. Safeguard's agent can open tested fix pull requests for vulnerable dependencies. It is off by default, every category starts on Manual, and you choose which categories run automatically. Safeguard helps prepare and keep evidence ready.
Try the framework mapping demo full screen.
Sources
- NIST, AI Risk Management Framework (AI RMF 1.0, NIST AI 100-1)
- NIST, AI RMF overview page
- NIST AIRC, AI RMF Playbook
Last reviewed: 18 October 2026.
