The NIST AI RMF requires no particular documents, because it is voluntary. In practice teams keep an AI policy, a roles record, an AI system inventory, risk assessments, test and monitoring plans, third party rules, a decommissioning procedure and incident records. Each maps to a Govern, Map, Measure or Manage subcategory.
Why documents at all?
The framework's outcomes read like statements that someone should be able to show are true: "policies are in place", "processes are followed and documented". A reader, whether a customer, auditor or your own board, will ask for proof. NIST says the Playbook is not a checklist and that organisations may borrow as few or as many suggestions as suit them, so choose a set that fits your size.
Govern documents
AI policy. GOVERN 1.2 expects the characteristics of trustworthy AI to be built into policies and procedures. NIST's list is valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy enhanced, and fair with harmful bias managed. Say what each means for your products in a few lines.
Roles and accountability. GOVERN 2.1 asks for clear roles, responsibilities and lines of communication. A one page RACI naming the executive owner, system owners, reviewers and who can pause a system is enough.
Risk management process and tolerance. GOVERN 1.3 and 1.4 cover processes and transparent risk management. Write how risk is rated, who accepts it, and what level needs executive approval.
AI system inventory. GOVERN 1.6 expects an inventory. Keep it current.
Decommissioning procedure. GOVERN 1.7 covers safe retirement.
Third party policy. GOVERN 6.1 addresses risks from third party software, data and other supply chain issues. Cover vendor due diligence, contract terms, data use limits and monitoring.
Culture and training. GOVERN 4.1 mentions a safety first mindset. Keep training records and review notes.
Map documents
System profile. For each system: intended purpose, users, context, known limits, data sources and legal constraints (MAP 1.1). A model card or readme can serve.
Categorisation and component record. MAP 2.1 asks for categorisation, and MAP 4.1 for documenting risks of components, including third party data or software.
Impact assessment. MAP 3.1 to 3.2 and 5.1 relate to benefits, costs and impacts. Record likelihood and magnitude for people and the organisation, with who was consulted.
Measure documents
Evaluation plan. MEASURE 1.1 asks for appropriate methods and metrics. List metrics, thresholds, test sets and tools, and why they fit the context.
Test results. MEASURE 2.x covers performance, safety, security, privacy and fairness. Keep dated reports, linked to the model or prompt version.
Monitoring plan. MEASURE 2.4 and 3.1 cover production behaviour and tracking risks over time.
Manage documents
Risk treatment plan. MANAGE 1.x asks that treatments for documented risks are prioritised and residual risk recorded.
Deactivation procedure. MANAGE 2.4.
Post deployment plan and incident log. MANAGE 4.1 and 4.3.
How long should each be?
Short. A page for the policy, a table for roles, a register for inventory and risks, and templates reused per system. Date them, name an owner and set a review cycle. NIST has said the AI RMF is being revised, so note the version you worked from.
What to do
- Write the one page AI policy using NIST's trustworthiness characteristics.
- Name roles and the person who can pause a system.
- Create the inventory and risk register.
- Make templates for system profile, impact assessment and test results.
- Write third party and decommissioning rules.
- Build the monitoring plan and incident log.
- Review everything on a fixed schedule.
Quick check
Does NIST require specific AI documents?
No. The framework is voluntary and outcome based. The documents above are a common way to show the outcomes are met.
Which version of the framework applies?
AI RMF 1.0 from January 2023. NIST says it is being revised, so record the version you used and check for updates.
How Safeguard helps automate this
Documents only help when they match what is running. 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. That leaves your team time to write the judgement calls, such as risk tolerance. Technical evidence comes from your real code and dependencies, using scans, SBOMs and reachability, which fits the component record and testing files. 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. Independent reviewers make their own assessment.
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.
