The NIST AI Risk Management Framework organizes AI risk management into four functions: Govern, Map, Measure, and Manage. This guide explains what each function actually requires for LLM systems, and where security testing fits into the framework.
The NIST AI Risk Management Framework is a voluntary framework for managing risks associated with AI systems throughout their lifecycle. Unlike the EU AI Act, it doesn't classify systems into binding risk tiers with legal penalties — instead it provides a structured methodology for identifying, measuring, and managing AI risk that organizations can apply regardless of jurisdiction.
Why it matters even though it's voluntary: NIST AI RMF is increasingly referenced in procurement requirements, vendor security reviews, and as the practical baseline enterprise customers expect when evaluating an AI vendor's risk management maturity.
Establishes organizational policies, accountability structures, and authorization processes for AI risk. In practice: every AI system and every test or scan run against it should link back to documented authorization, creating an accountable audit trail rather than ad hoc testing with no ownership.
Identifies the context and risk surface specific to a given AI system — what it does, what data it touches, and what attack surfaces it exposes across input, retrieval, model, and output layers. This is where you systematically map your specific threat landscape rather than applying generic risk assumptions.
Evaluates the mapped risks through systematic testing and evaluation. This is where security testing does its primary work — producing technical evidence of how the system actually behaves under adversarial conditions, not just documented intent. Findings should be exportable in standard formats for audit and compliance documentation.
Applies remediation and ongoing risk treatment based on what Measure surfaces. Each finding should map to specific technical controls, with the ability to re-test continuously as the model, prompts, or deployment change over time.
Security testing is the primary mechanism for satisfying the Measure function — it's how an organization gets actual evidence of AI system behavior rather than relying on documented policy alone. The output of testing then feeds directly into Manage, where findings get translated into specific remediation actions and control changes.
Treating NIST AI RMF as a paperwork exercise — writing governance documents without actually measuring system behavior — defeats the purpose of the framework. Govern and Map without rigorous Measure is unverified intent; Measure without Govern and Map is testing without context or accountability. All four functions need to operate together.
The HexTyx AI Security Assessment produces structured testing evidence that maps directly to the Measure function, with findings ready to feed your Manage process.