Skip to content

Generative AI risk assessment

Questions for assessing one generative AI system, grouped by the NIST AI RMF functions Govern, Map, Measure, and Manage, with answer, evidence, status, and owner per question.

v1.0CSVCurrent as of November 18, 2026No signup, generated in your browser

This assessment follows the structure of the NIST AI Risk Management Framework (AI RMF 1.0): every question sits under one of its four Core functions, Govern, Map, Measure, and Manage, and names the AI RMF category it serves. Where a question addresses a risk specific to generative AI, it is tagged with that risk's name from the NIST Generative AI Profile (NIST AI 600-1). The questions are our own; the structure and risk names are NIST's, and the framework documents linked below remain the authority. Use one file per system or use case.

Preview

The first 6 of 21 rows, rendered from the same data the download is built from.

RefAI RMF functionAI RMF categoryQuestionGAI Profile riskAnswerEvidenceStatus (Met/Partial/Not met/Not applicable)OwnerFollow-up
G1GovernGOVERN 1Is there a written policy stating which generative AI uses are permitted, which need approval, and which are prohibited?Information Security; Data Privacyblankblankblankblankblank
G2GovernGOVERN 2Is a named business owner accountable for this system, separate from the team that built or bought it?blankblankblankblankblankblank
G3GovernGOVERN 2Who can approve, pause, or retire this system, and is that authority written down?blankblankblankblankblankblank
G4GovernGOVERN 4Can staff report a wrong or harmful output, and does someone read and act on those reports?Confabulationblankblankblankblankblank
G5GovernGOVERN 6Have the model provider's terms on retention of your inputs and their use for training been reviewed and recorded?Data Privacy; Intellectual Propertyblankblankblankblankblank
G6GovernGOVERN 6Is there a plan for the provider changing, deprecating, or withdrawing the model you depend on?Value Chain and Component Integrationblankblankblankblankblank

Formulas: Add a COUNTIFS summary by function and status so each function shows how many questions are Met, Partial, and Not met. Put any Not met under Govern first in the summary, because the other three functions assume an owner and a policy exist.

How to use it

  1. Assess one system or use case per file. A chatbot answering from public documentation and an agent with write access to a production system have different answers to almost every question.
  2. Work through Map early, since it decides which Measure questions apply. Govern still frames the whole file: the AI RMF describes it as a cross-cutting function.
  3. Put a link or document name in Evidence for every answer. An answer with no evidence is Partial at most.
  4. Mark Not applicable only with a reason in Follow-up, so a reviewer can tell a considered exclusion from a skipped question.
  5. Send every Not met that will not be fixed before launch to the risk acceptance record, with an owner and an expiry.

Assessing a generative AI use case

Framework references

The template follows the structure of these documents and cites them; it does not reproduce them. Read the originals for the authoritative text.

Changelog

v1.0 (November 18, 2026):
First release. Twenty-one questions across the four AI RMF functions, tagged to NIST AI 600-1 risk names.

Common questions

What are the four functions of the NIST AI RMF?
Govern, Map, Measure, and Manage. NIST describes Govern as cross-cutting: it sets the policies, accountability, and culture the other three depend on. Map establishes the context of a system and frames its risks, Measure analyzes and tracks those risks, and Manage prioritizes and acts on them. Every question in the template belongs to one function and one category.
What is the NIST Generative AI Profile?
NIST AI 600-1, published in July 2024, is a companion profile to the AI RMF that applies the framework to generative AI. The profile names twelve risks that generative AI creates or makes worse, among them confabulation, data privacy, information security, and value chain and component integration. The template tags questions with those risk names where they apply.
Does completing a NIST AI RMF assessment make a system compliant?
No. The AI RMF is voluntary guidance with no certification, so there is no formal compliance to achieve. The assessment records, with evidence, how one system is governed, mapped, measured, and managed. That record supports internal approval, audits, and customer questions, and it does not replace legal review of regulation such as the EU AI Act.

Filed under AI Governance.