NIST AI RMF 600-1 GenAI Profile

What it is #

NIST AI 600-1 (July 2024) is the Generative AI Profile of the AI RMF. It identifies twelve GenAI-specific risk categories that don’t appear in the core framework — risks that emerge when an AI system can generate text, code, or actions rather than just classify or score.

The twelve risks: confabulation (hallucination), dangerous-content, data privacy, environmental, human-AI configuration, information integrity, information security, intellectual property, harmful bias, obscenity, value chain (third-party model risk), and CBRN (chemical/ biological/radiological/nuclear).

How Regulus maps to it #

Five of the twelve GenAI risks have direct runtime evidence in Regulus:

GenAI risk	Regulus control
Confabulation (hallucination)	Model-risk plugin tier gate + audit envelope model_confidence field
Data privacy	Privacy plugin (BeforeModelCallback redaction)
Information integrity	Audit plugin hash chain + AfterModelCallback re-redaction
Information security	Identity expiry guard + dual-control kill switch
Value chain	Model registry + GRC adapter dispatch on every tier-3 invocation

Citations land as nist-ai-rmf-600-1:DataPrivacy etc. on every relevant event.

The other seven risks (CBRN, harmful bias, IP, obscenity, environmental, human-AI configuration, dangerous-content) are outside the runtime plane — they’re addressed by your organisational governance processes, your model training pipeline, your dataset curation, and your usage policy. Regulus doesn’t claim runtime coverage for risks that aren’t runtime problems.

Activating #

regulus:
  frameworks:
    - nist-ai-rmf
    - nist-ai-rmf-600-1

Activate both — the GenAI profile depends on the base AI RMF.

Cross-references #

Most EU AI Act Article 9 risk-management requirements map to a combination of AI RMF core + 600-1. The Regulus EU AI Act profile auto-activates both NIST frameworks when enabled.