Skip to main content

Governance with the NIST AI RMF

The NIST Artificial Intelligence Risk Management Framework (AI RMF 1.0), published January 2023, is voluntary U.S. government guidance for organizations that design, develop, deploy, or use AI systems. The full text is NIST.AI.100-1. NIST also publishes an AI RMF Playbook with suggested actions.

This page does not replace the source documents; it shows how CSN readers can align LLM product work (see Risk landscape) with a governance structure auditors recognize.

The four core functions

NIST organizes AI risk activities into four functions:

FunctionPurposeExamples for LLM products
GovernCulture, policies, accountabilityAI security policy; who approves model changes; vendor rules for foundation models.
MapContext and known risksData flows for prompts; tool/MCP inventory; user populations; abuse scenarios.
MeasureAssess and track riskEvaluations for jailbreaks; leakage tests; cost/DoS monitoring; bias/safety metrics where required.
ManagePrioritize and mitigateHuman review for high-risk tools; rate limits; incident playbooks; patching supply-chain deps.

Govern is cross-cutting: it should inform how Map, Measure, and Manage are run, not be a one-time checklist.

Trustworthiness characteristics

NIST highlights characteristics to consider across the lifecycle, including security and resilience, privacy, accountability and transparency, safety, and fairness with managed bias. For security-focused readers, the point is simple: security is one dimension; product and legal stakeholders may prioritize others in parallel.

Mapping OWASP LLM risks to RMF activities

Rough alignment (not official NIST text):

  • Map — Identify which OWASP LLM items apply per feature; document trust boundaries (Threat modeling).
  • Measure — Run structured tests (prompt injection suites, tool-abuse cases, retrieval attacks, token abuse).
  • Manage — Implement controls from AI Automation (tool narrowing, sandboxing, n8n hardening) and organizational policy.

Practical takeaway

Use the RMF as a conversation framework with leadership and GRC: “We mapped these LLM risks, measure them with these tests, and manage them with these controls and owners.” That pairs engineering work with defensible documentation.

References