The NIST AI Risk Management Framework is voluntary and offers no certificate, so there is no official auditor and no fixed evidence list. A customer or assessor will ask what your Govern, Map, Measure and Manage functions produce. Expect documents on roles, intended use, test results, monitoring and incident handling. Software can help gather them but cannot attest for you.
What is the AI RMF?
NIST released AI RMF 1.0 on 26 January 2023 and describes it as intended for voluntary use. A Playbook accompanies it, and a Generative AI Profile (NIST-AI-600-1) came out on 26 July 2024. NIST's page also says the framework is being revised as part of the White House AI Action Plan, so check for updates.
The core has four functions. Govern is cross-cutting and informs the other three. Map sets context. Measure analyzes, tests and tracks risks. Manage prioritizes and responds. NIST says subcategories are not a checklist, which is why an assessor's list depends on who is asking.
What would show up for Govern?
NIST's core page gives these examples.
- Legal and regulatory requirements for AI understood, managed and documented (Govern 1.1).
- Roles, responsibilities and lines of communication for AI risk documented and clear (Govern 2.1).
- Teams document the risks and potential impacts of the AI technology they work with (Govern 4.2).
Evidence: an AI policy, a register of applicable laws, an owner list for each AI system, and a risk register with entries for each system.
What would show up for Map?
- Intended purposes and prospective deployment settings understood and documented (Map 1.1), with documented participation of interdisciplinary people (Map 1.2).
- Documentation that gives enough information to relevant AI actors, including the system's knowledge limits and how humans oversee its output (Map 2.2).
Evidence: a system description per model or feature, an intended use statement, known limits, and records of who reviewed the context.
What would show up for Measure?
- Test sets, metrics and details about the tools used for testing are documented (Measure 2.1).
- The AI model is explained, validated and documented (Measure 2.9).
- Approaches, people and documentation exist for regularly identifying and tracking existing and emerging risks (Measure 3.1).
Evidence: your evaluation sets, results over time, the tools and versions you used, and a log of risks found since launch.
What would show up for Manage?
- Response plans for high-priority risks developed and documented, with options such as mitigating, transferring, avoiding or accepting (Manage 1.3).
- Processes for tracking, responding to and recovering from incidents and errors are followed and documented (Manage 4.3).
Evidence: risk treatment decisions with owners and dates, an incident log, and postmortems.
How does this tie to security work?
The generative AI profile and the core both depend on the basics: inventory of AI components, access control and testing. A software inventory, a record of scans and a map from controls to frameworks can supply part of the paper trail. Safeguard offers compliance automation and mapping, and it does not make you NIST AI RMF compliant, because the framework has no certification.
What to do
- Ask your customer or assessor which framework version and which subcategories they care about.
- Build an inventory of AI systems with owner, purpose and risk rating.
- Write the intended use and limits for each system.
- Store test sets and results with dates, and rerun on each model change.
- Keep an AI incident log and review it monthly.
- Record each risk decision with a name and a date.
Quick check
Is there a NIST AI RMF certification?
No. NIST describes the framework as voluntary. Any proof is what you document and what a customer or assessor accepts.
Is the Generative AI Profile separate from the framework?
It is a companion document, NIST-AI-600-1, released on 26 July 2024, that helps identify risks unique to generative AI.
Sources
Last reviewed: 18 October 2026.
