Login Get a Demo
01 Glossary · Governance and risk

What Is AI Risk Management?

The discipline behind every AI control: how the cycle runs, what NIST, ISO and the EU AI Act each ask for, and where the legal duty starts.

The short answer

What is AI risk management?

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

AI risk management is the continuous practice of identifying, assessing, treating and monitoring the risks an AI system poses to people, to the organization and to society, across its whole lifecycle.

It adapts enterprise risk management to systems that are probabilistic, that change after deployment and that often arrive from third parties.

Key takeaways
  • NIST AI RMF 1.0 organizes the work into four functions: GOVERN, MAP, MEASURE and MANAGE.
  • The NIST framework is voluntary. A risk management system is a legal duty for providers of high-risk AI under EU AI Act Article 9.
  • NIST names four responses to a risk: mitigating, transferring, avoiding or accepting (MANAGE 1.3).
  • No framework sets your risk tolerance. NIST says the AI RMF “does not prescribe” it.
By the numbers Each figure links to its sourceAll sources 
02 The cycle

How does AI risk management work?

The four NIST AI RMF functions describe the cycle. NIST notes that its outcomes are not a checklist, nor necessarily an ordered set of steps.

  1. Govern

    Set policy, roles and risk tolerance, and keep an inventory of AI systems (GOVERN 1.6). NIST calls this function cross-cutting: it informs the other three.

  2. Map

    Establish context for each system: intended purpose, the people affected, legal requirements, third-party components, and the likelihood and magnitude of each impact (MAP 1.1, MAP 5.1).

  3. Measure

    Test and track the mapped risks with quantitative and qualitative methods, covering validity, safety, security, privacy and bias (MEASURE 2).

  4. Manage

    Prioritize treatment by impact, likelihood and available resources (MANAGE 1.2), document residual risk, and monitor after deployment (MANAGE 4.1).

03 Three texts

NIST AI RMF vs. ISO/IEC 23894 vs. EU AI Act Article 9

Two are voluntary and one is law. Read them together: the frameworks say how, the Act says who must.

NIST AI RMF 1.0ISO/IEC 23894:2023EU AI Act, Article 9
StatusVoluntary US frameworkVoluntary international guidanceBinding EU law
PublishedJanuary 26, 2023; NIST says a revision is under wayFebruary 6, 2023In force since August 1, 2024
Applies toAny organization that adopts itOrganizations that develop, produce, deploy or use AIProviders of high-risk AI systems
ShapeFour functions and 72 subcategoriesA process: risk criteria, identification, analysis, treatment, monitoring and reviewA documented, continuous and iterative process across the lifecycle
04 What is different

Why does AI need its own approach?

Section 1.2 of the NIST framework names four challenges that general risk practice does not settle.

01

Measurement

NIST’s first challenge: AI risks that are not well defined or adequately understood are difficult to measure, in numbers or in words (section 1.2.1).

02

Tolerance

How much risk is acceptable depends on context, law and sector norms. MAP 1.5 asks you to determine and document it (section 1.2.2).

03

Prioritization

NIST warns that trying to eliminate negative risk entirely can be counterproductive. Resources go to the most material risks first (section 1.2.3).

04

Integration

AI risks “should not be considered in isolation”. They belong beside cybersecurity and privacy in enterprise risk management (section 1.2.4).

05 Keep going

Related reading.

06 Questions

Frequently asked questions.

Straight answers, kept current. Last updated .

Ask us anything →
Is AI risk management legally required?

It depends on your role and the system. The NIST AI RMF and ISO/IEC 23894 are voluntary. Under the EU AI Act, providers of high-risk AI systems must establish, implement, document and maintain a risk management system (Article 9), and deployers must monitor operation under Article 26(5). The Commission’s timeline applies the Annex III high-risk rules from December 2, 2027.

What is the difference between AI risk management and an AI risk management platform?

AI risk management is the discipline: the people, policies and process that decide how much AI risk to take and how to treat it. A platform is software that supports the discipline by discovering AI systems, recording assessments and producing evidence. You can practice the discipline without the software, but the software does nothing useful without the discipline.

Who owns AI risk management?

NIST places the decision with the top of the organization: under GOVERN 2.3, executive leadership takes responsibility for decisions about AI risks, and GOVERN 2.1 asks that roles and lines of communication for mapping, measuring and managing those risks be documented. Day to day, the work is shared by risk, security, legal, data and product teams.

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.