The NIST AI RMF has no certification or formal audit, so surviving an assessment means being ready for customer, board or regulator reviews with records that show you govern, map, measure and manage your AI systems. Keep an AI inventory, risk assessments, test results and decisions, all dated and owned.
What is the AI RMF?
NIST released AI RMF 1.0 on 26 January 2023. It is described as voluntary, organized around four functions (Govern, Map, Measure and Manage), and accompanied by a Playbook. NIST released a Generative AI profile, NIST-AI-600-1, on 26 July 2024. NIST's page says AI RMF 1.0 is being revised, so check the current version before building a program around it.
Who will ask for evidence?
Enterprise customers in security reviews, procurement teams, investors in due diligence, internal audit, and sometimes regulators who use the RMF as a reference. They typically want to see that you have a repeatable process, not a perfect record.
What evidence supports each function?
Govern. Policy for AI use and development, named owners and an escalation path, risk appetite statement, training records, and records of leadership review.
Map. An inventory of AI systems with purpose, users, data sources, models and suppliers; context and intended use statements; assessments of potential harms and affected groups; notes on legal and contractual constraints.
Measure. Test plans and results for accuracy, resilience, bias where relevant, security and privacy; evaluation of generative systems for the risks in the NIST generative profile; monitoring metrics and thresholds.
Manage. Risk treatment decisions, with owners and dates; release approvals; incident log and response records; decommissioning decisions; supplier reviews and change tracking.
Build a binder customers can read
Create one page per AI system: purpose, owner, data, model and version, risks, tests run with dates, approvals, monitoring, and incidents. Link each item to the artifact. A reviewer who can read a page in ten minutes will trust you more than one who gets a hundred page policy.
How do you rehearse?
Pick a system and ask a colleague to review it as a customer would. Questions to expect: How do you know it works? What happens when it is wrong? Who can switch it off? What data trained it? What changed since the last review? Time how long it takes to answer each with an artifact.
How long should evidence be kept?
NIST does not set a retention period for the RMF, so decide it yourself from your contracts, sector rules and the life of the system. A practical rule is to keep records for at least as long as the system runs and for the period customers may ask about it afterward. Write the rule into your policy, store the records somewhere with access control and history, and make sure the model version, data version and test date appear on every record, so a reviewer can tell exactly which release a result describes.
Mistakes to avoid
- Claiming alignment without records.
- Ignoring AI features inside vendor products.
- Testing once at launch and never again.
- No owner for model changes.
- Writing a policy that bans uses people already practice, creating hidden shadow use.
What to do
- Appoint an AI risk owner and publish a short policy.
- Build and maintain an AI system inventory.
- Write a one page profile per system with risks and intended use.
- Run and record tests, and set monitoring thresholds.
- Log incidents and decisions with owners and dates.
- Review suppliers and model changes at set intervals.
- Hold a mock customer review each year.
Quick check
Is there a NIST AI RMF certification?
No. NIST describes the framework as voluntary. Customers or auditors may review your practices, but there is no NIST certificate.
What is NIST-AI-600-1?
The Generative AI profile for the AI RMF, released by NIST on 26 July 2024 to help identify risks specific to generative AI.
How Safeguard helps automate this
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 audits. For AI systems this keeps inventory, tests and review records current instead of frozen at launch.
Technical evidence comes from your real code and dependencies, through scans, SBOMs and reachability, which shows the software supply chain behind your AI features. Safeguard's agent can open tested fix pull requests for vulnerable dependencies and other findings. It is off by default, every category starts on Manual, and you choose which categories run automatically and set the approval level. The AI RMF has no certifier, and any independent review is done by the party you choose. Safeguard helps prepare and keep evidence ready, and does not certify or guarantee an outcome.
Want to see it on your own stack? Book a call.
Sources
Last reviewed: 18 October 2026.
