Login Get a Demo
01 Glossary · Governance and risk

What Is an AI Governance Framework?

The policies, roles and controls that decide how AI gets approved, used and reviewed, with the NIST subcategory and ISO clause behind each part.

The short answer

What is an AI governance framework?

Last updated
Read time
4 min read
From
Cranium AI
Share on LinkedIn

An AI governance framework is the documented set of policies, roles, processes and controls an organization uses to decide which AI it will build, buy and run, and on what conditions.

The term also covers the published models companies adopt as a template, such as the NIST AI RMF and ISO/IEC 42001.

Key takeaways
  • A framework is a structure of decisions and owners. Software can run it but cannot stand in for it.
  • The NIST AI RMF is voluntary and ISO/IEC 42001 is certifiable. Of the three, only the EU AI Act is law.
  • NIST’s GOVERN function alone holds 19 subcategories, from policy and inventory to third-party risk.
  • Seven parts recur: policy, roles, inventory, risk assessment, impact assessment, third parties and review.
By the numbers Each figure links to its sourceAll sources 
  • 72

    subcategories in NIST AI RMF 1.0, grouped into 19 categories and four functions

    Source: NIST, AI RMF Core

  • 19

    of those subcategories sit in GOVERN, the function that covers policy, roles and inventory

    Source: NIST, AI RMF Core

  • 4

    functions organize the framework: GOVERN, MAP, MEASURE and MANAGE

    Source: NIST, AI RMF Core

02 Three meanings

One term, three things

People use the phrase for the program they run, the model they copied and the law they answer to. Keeping them apart avoids most confusion.

01

Your internal framework

The operating model you write: an AI policy, an owner with authority, an inventory, assessment steps and a review cycle. It reflects your own risk tolerance, which NIST says its framework “does not prescribe”.

NIST AI RMF 1.0
02

A published framework

A reference model you adopt instead of inventing one. The NIST AI RMF sets outcomes under four functions. ISO/IEC 42001 sets requirements for an AI management system, and ISO/IEC 42006 governs the bodies that certify it.

Frameworks compared
03

The legal baseline

Law is not a framework you choose. The EU AI Act binds providers and deployers of AI in the EU, so whichever model you follow has to produce the evidence the Act asks for.

The EU AI Act
03 Building blocks

What does an AI governance framework contain?

Seven components recur. Each row names the NIST AI RMF subcategory and the ISO/IEC 42001 clause that address it. Clause titles are as listed in NIST’s crosswalk to the standard.

ComponentNIST AI RMFISO/IEC 42001
AI policyGOVERN 1.2Clause 5.2, AI policy
Roles and accountabilityGOVERN 2.1 and 2.3Clause 5.3, roles, responsibilities and authorities
AI inventoryGOVERN 1.6Annex B.4.2, resource documentation
Risk assessment and treatmentGOVERN 1.3 and MANAGE 1.2Clauses 6.1.2 and 6.1.3
Impact assessmentGOVERN 4.2Clause 6.1.4, AI system impact assessment
Third-party AIGOVERN 6.1Annex B.10.3, suppliers
Monitoring and reviewGOVERN 1.5Clause 9.1, monitoring, measurement, analysis and evaluation
04 How to build

How do you build one?

  1. Name the owner

    Give one executive accountability and a cross-functional group the authority to approve, restrict or stop AI use. GOVERN 2.3 places AI risk decisions with executive leadership.

  2. Write the policy and the tolerance

    State which AI is allowed, for what and on which data. Document your risk tolerance (MAP 1.5), because no framework sets it for you.

  3. Inventory what exists

    Record every model, agent and vendor AI system before adding rules. Controls can only be scoped against a list.

  4. Assess by tier

    Route each use through risk assessment, and impact assessment where people are affected, with depth set by risk tier.

  5. Review on a schedule

    Monitor, audit and report to management, and revisit an entry when the system, the vendor or the law changes.

05 Keep going

Related reading.

06 Questions

Frequently asked questions.

Straight answers, kept current. Last updated .

Ask us anything →
Is an AI governance framework legally required?

Not by that name. The NIST AI RMF is voluntary and ISO/IEC 42001 certification is a choice. The EU AI Act is different: it imposes binding duties, including a documented risk management system for providers of high-risk AI systems under Article 9. The practical reason to adopt one is that customers, auditors and regulators ask for proof that it exists.

What is the difference between an AI governance framework and an AI governance platform?

The framework is the set of decisions: the policy, the owners, the risk tiers and the evidence each tier requires. The platform is software that carries those decisions out by keeping the inventory, recording assessments and mapping evidence to frameworks. A platform with no framework behind it automates a process nobody defined.

Which AI governance framework should we start with?

Start with the NIST AI RMF if you want a free, detailed vocabulary: it has 72 subcategories and a companion Playbook of suggested actions. Choose ISO/IEC 42001 if you need a management system an auditor can certify. If you sell or deploy AI in the EU, map either one to the EU AI Act, because only the Act is law.

07 See it for yourself

From the definition
to the evidence.

See how Cranium AI discovers, observes, governs, secures and proves every AI model, agent and vendor in your enterprise.