ISO 42001: AI Management System Requirements and Certification

ISO 42001 (formally ISO/IEC 42001:2023) is the first international management system standard for artificial intelligence. It does for AI what ISO/IEC 27001 does for information security: it sets out how an organisation governs AI through policy, risk assessment, controls, monitoring and improvement, in a form that an accredited body can audit and certify.

For software companies adding AI to their products, ISO/IEC 42001 answers a question customers are now asking in procurement: how do you make sure your AI is reliable, fair, secure and used as intended? It also gives a structure for the obligations arriving in AI laws such as the EU AI Act. This guide explains the clauses, the AI-specific requirements, the Annex A controls, documentation, certification, costs and the mistakes to avoid.

What is ISO/IEC 42001?

ISO/IEC 42001 specifies requirements for establishing, implementing, maintaining and continually improving an artificial intelligence management system (AIMS) within an organisation. It applies to any organisation that provides or uses products or services that use AI systems, whatever its size, type or sector.

The standard is written by ISO/IEC JTC 1/SC 42, the joint committee on artificial intelligence. It uses the harmonised structure common to ISO management system standards, so its clauses 4 to 10 mirror those of ISO/IEC 27001 and ISO 9001. It adds four annexes:

• Annex A (normative): reference control objectives and controls.
• Annex B (normative): implementation guidance for the Annex A controls.
• Annex C (informative): potential AI-related organisational objectives and risk sources.
• Annex D (informative): use of the AI management system across domains or sectors.

Terms such as "AI system" and "AI stakeholder roles" come from ISO/IEC 22989, and risk management guidance comes from ISO/IEC 23894.

History and why ISO 42001 was introduced

SC 42 was created in 2017 to develop AI standards. As AI moved from research into products, organisations needed a way to show that AI was developed and used responsibly, and regulators began drafting AI laws. Ethical principles (from the OECD, UNESCO and others) said what to aim for but not how to manage it. ISO/IEC 42001 was published in December 2023 to fill that gap with an auditable management system.

Its purpose is to help organisations develop, provide or use AI systems responsibly in pursuing their objectives and meeting applicable requirements, obligations to interested parties and their expectations. Because it follows the common management system structure, organisations with ISO/IEC 27001 or ISO 9001 can extend existing processes rather than build a separate one.

Who needs ISO 42001?

No law requires it. It is useful for:

• AI product and platform vendors, including companies offering models or AI APIs.
• SaaS and software companies with AI features in their products.
• IT service firms that build AI solutions for clients.
• Enterprises using AI in decisions about customers or employees, such as banks, insurers, HR and healthcare providers.
• Any organisation that wants independent evidence of AI governance for customers, investors or regulators.

The standard is global and sector-neutral. Its value grows where AI law applies: under the EU AI Act, providers of high-risk systems need a quality management system and risk management, and an AIMS gives that a structure. In the United States, Colorado's original 2024 AI Act named ISO/IEC 42001 and the NIST AI RMF as examples of recognised risk management frameworks; it was replaced in 2026 by a narrower Colorado automated decision law, but the standard remains good evidence of responsible AI practice.

Key requirements: clauses 4 to 10

Clause 4: Context

Understand the organisation and its context, including its roles in relation to AI (for example provider, producer, customer, partner or user of AI), the intended purpose of the AI systems, the needs of interested parties, and define the scope of the AIMS. Clause 4.1 also asks whether climate change is relevant, as with all ISO management system standards after the 2024 amendments.

Clause 5: Leadership

Top management commitment, an AI policy (5.2) that fits the organisation's purpose and provides a framework for AI objectives, and assigned roles and responsibilities.

Clause 6: Planning

• AI risk assessment (6.1.2): identify, analyse and evaluate risks that could prevent achieving AI objectives, using defined criteria.
• AI risk treatment (6.1.3): choose treatment options and controls, compare them with Annex A, and produce a Statement of Applicability justifying inclusions and exclusions.
• AI system impact assessment (6.1.4): assess the potential consequences for individuals, groups of individuals and societies of the development, provision or use of AI systems. This is the requirement with no direct equivalent in ISO/IEC 27001.
• AI objectives (6.2) and planning of changes (6.3).

Clause 7: Support

Resources, competence, awareness, communication and documented information. Resources include data, tooling, computing and people with the right expertise.

Clause 8: Operation

Operational planning and control, performing AI risk assessments, AI risk treatment and AI system impact assessments at planned intervals or when significant changes occur.

Clause 9: Performance evaluation

Monitoring and measurement, internal audit and management review.

Clause 10: Improvement

Continual improvement, and nonconformity and corrective action.

Annex A controls explained

Annex A lists 38 controls under nine control objective areas. As with ISO/IEC 27001, you determine which controls are necessary through risk treatment and justify the rest in the Statement of Applicability.

Area	What the controls address
A.2 Policies related to AI	AI policy, alignment with other policies, review
A.3 Internal organisation	AI roles and responsibilities, reporting of concerns
A.4 Resources for AI systems	Documenting data, tooling, system and computing resources and human resources
A.5 Assessing impacts of AI systems	Impact assessment process and documentation, impacts on individuals, groups and society
A.6 AI system life cycle	Objectives for responsible development, requirements, design, verification and validation, deployment, operation and monitoring, technical documentation, event logs
A.7 Data for AI systems	Data for development and enhancement, acquisition, quality, provenance and preparation
A.8 Information for interested parties	System documentation and information for users, external reporting, communication of incidents
A.9 Use of AI systems	Processes for responsible use, objectives for responsible use, intended use
A.10 Third-party and customer relationships	Allocating responsibilities, suppliers, customers

For a software company, the life cycle (A.6) and data (A.7) controls carry the most engineering work: documented requirements for each AI feature, evaluation before release, logging during operation, version control for models and data, and clear provenance of training and evaluation data.

How to implement ISO 42001 step by step

1. Decide scope and roles. Which AI systems, products, teams and locations are in scope, and whether you are a provider, developer or user of each system.
2. Build an AI inventory. Every AI system and model, including third-party AI services, with purpose, owner, data and users.
3. Write the AI policy and assign roles: an AIMS owner, system owners, and a route to raise concerns.
4. Define risk criteria and run AI risk assessments for each system, covering accuracy, robustness, bias and fairness, security, privacy, transparency, misuse and dependency on suppliers. Annex C lists risk sources to consider.
5. Run AI system impact assessments for effects on individuals, groups and society, scaled to the system's potential for harm.
6. Treat risks: choose controls, compare with Annex A, write the Statement of Applicability and a treatment plan.
7. Implement life cycle controls: requirements and acceptance criteria per AI feature, evaluation datasets and metrics, human oversight where needed, release approvals, monitoring of model performance and drift, event logging, and change management for models and prompts.
8. Manage data: provenance, quality checks, consent and licensing, retention.
9. Manage suppliers and customers: document responsibilities with model providers and customers, and give customers the information they need to use your AI correctly.
10. Train people on the policy and their responsibilities.
11. Audit and review: internal audit and management review, then corrective actions.
12. Book certification with an accredited body.

Documentation and records required

• Scope of the AIMS (4.3).
• AI policy (5.2).
• AI risk assessment process and results (6.1.2, 8.2).
• AI risk treatment process, results and Statement of Applicability (6.1.3, 8.3).
• AI system impact assessment process and results (6.1.4, 8.4).
• AI objectives (6.2).
• Evidence of competence (7.2).
• Operational documentation needed to have confidence processes run as planned (8.1).
• Monitoring results, internal audit programme and results, management review results (9.1 to 9.3).
• Nonconformities and corrective actions (10.2).
• Annex A records as applicable: AI system documentation, data provenance and quality records, verification and validation records, event logs, user information, supplier agreements.

Audit and certification

Certification follows the same pattern as other ISO management system standards: a Stage 1 documentation and readiness review, a Stage 2 audit of implementation and effectiveness, a three-year certificate with surveillance audits in the following two years, then recertification. Choose a certification body accredited for ISO/IEC 42001 by a national accreditation body that belongs to the international accreditation arrangement now run by the Global Accreditation Cooperation (which replaced the IAF and ILAC in January 2026); ISO/IEC 42006 sets the additional requirements those bodies must meet. Ask for accreditation evidence, because the scheme is new and not every body offering "42001 certificates" is accredited for it. Accredited certificates can be checked on IAF CertSearch.

Auditors will expect to see AI systems that have actually gone through the risk and impact assessment and life cycle controls, not only policies. Plan for at least one full cycle of operation, internal audit and management review before Stage 2.

Timeline and cost considerations

Organisations that already hold ISO/IEC 27001 can reuse their management system processes (document control, internal audit, management review, corrective action) and usually move faster. The time goes into the AI inventory, the risk and impact assessments for each system, and the life cycle controls. The number and complexity of AI systems in scope is the main driver.

Cost factors: certification body fees (audit days), internal staff time across product, engineering, data, legal and leadership, tooling for model evaluation, monitoring, logging and documentation, and any external consultants or internal auditors.

Benefits and the risk of not having it

ISO/IEC 42001 gives a credible, independent answer to customers' AI governance questions, prepares the organisation for AI laws by building inventories, risk assessments, documentation and oversight that those laws require, and reduces the chance of AI failures that reach customers. Because it shares a structure with ISO/IEC 27001 and ISO 9001, it can be added to an existing integrated management system.

There is no legal penalty for not being certified. The risks of having no AI governance are practical: AI features shipped without evaluation or monitoring, unclear responsibility when something goes wrong, and procurement teams that will not approve AI vendors without evidence.

Common mistakes and challenges for IT companies

• An incomplete AI inventory. Third-party AI services, SDKs and internal tools are often missed.
• Treating the impact assessment as a form. It should change design decisions, such as adding human review or restricting use.
• No evaluation baseline. Without defined metrics and test data, you cannot show a release is fit for purpose or detect drift.
• Ignoring prompt and model changes. Changing a prompt or swapping a model version can change behaviour as much as code changes; put them under change management.
• Unclear supplier responsibilities. If you build on a third-party model, document what the model provider does and what you do.
• Using a non-accredited certifier. The certificate may not be accepted by customers.
• Assuming certification equals legal compliance. It supports compliance with AI laws but does not replace their specific requirements.

Best practices, tools and technologies

• Keep model cards and data sheets for every AI system; they feed technical documentation and user information.
• Version models, prompts, datasets and evaluation suites together, and record which combination is in production.
• Automate evaluation in the release pipeline: accuracy, robustness, safety and fairness checks with thresholds.
• Log inputs, outputs, model versions and human interventions with access control and retention rules.
• Monitor production for drift, error rates, abuse and user feedback.
• Build human review and override into features that affect people.
• Integrate the AIMS with ISO/IEC 27001 for security and ISO/IEC 27701 for privacy.

These engineering controls are part of the AI features we build in custom software projects.

How ISO 42001 relates to other frameworks

Framework	Relationship
EU AI Act	Law. ISO/IEC 42001 can structure the quality management and risk processes the Act requires, but does not by itself give a presumption of conformity.
NIST AI RMF	Voluntary US framework with govern, map, measure and manage functions. Compatible; NIST has published crosswalks.
ISO/IEC 23894	Guidance on AI risk management that supports clause 6.1.
ISO/IEC 27001	Same structure; many organisations run an integrated system.
ISO/IEC 27701	Privacy management for personal data used in AI.
Colorado AI Act	The original 2024 law named ISO/IEC 42001 as a recognised framework; the 2026 replacement focuses on notices and explanations.

Real-world implementation examples

Several large cloud, software and AI providers have announced accredited ISO/IEC 42001 certifications for AI services and publish the certificates in their compliance portals, alongside their ISO/IEC 27001 certificates. Reading their scope statements shows how they define the AI systems and roles covered.

An illustrative case for a mid-sized SaaS company with AI features: it already holds ISO/IEC 27001. It adds an AI inventory (five features, two using a third-party model API), runs risk and impact assessments that lead it to add human review to one feature that drafts customer-facing messages, creates evaluation suites for each feature, and puts prompts and model versions under change control. Its existing internal audit and management review cycle absorbs the AIMS, and the certification audit is run as an extension of the ISO/IEC 27001 surveillance audit.

Next steps

1. Build your AI inventory and decide your role for each system.
2. Write a short AI policy and assign owners.
3. Run risk and impact assessments on your highest-impact AI features first.
4. Put evaluation, logging and change control around those features.
5. Select an accredited certification body and plan the audit.

Talk to us about the engineering controls.

Frequently asked questions

Is ISO 42001 mandatory?

No. It is a voluntary standard. Customers, partners and investors may ask for it, and it can support compliance with AI laws.

What is an AI management system?

The policies, processes, roles and controls an organisation uses to govern the development, provision or use of AI systems, and to monitor and improve them.

How many controls are in ISO 42001 Annex A?

38 controls grouped under nine areas, from AI policies to third-party and customer relationships. You select those needed through risk treatment.

Is ISO 42001 only for AI developers?

No. It applies to organisations that develop, provide or use AI. A company using third-party AI in its operations can implement it too.

Does ISO 42001 certification mean EU AI Act compliance?

No. It helps with management, risk and documentation processes, but the AI Act has specific requirements, conformity assessments and deadlines that must be met separately.

What is an AI system impact assessment?

An assessment of the potential consequences of an AI system for individuals, groups and society, required by clause 6.1.4 and supported by Annex A.5 controls.

Can we combine ISO 42001 with ISO 27001?

Yes. They share the same clause structure, so document control, audits, management review and corrective action can be integrated, and audits can be combined.

How long is an ISO 42001 certificate valid?

Three years, with surveillance audits in the first and second years and a recertification audit before expiry, as with other ISO management system certifications.

What is ISO/IEC 42006?

The standard setting additional requirements for bodies that audit and certify AI management systems against ISO/IEC 42001, which helps make certificates comparable.

Do we need ISO 42001 if we only use ChatGPT or similar tools internally?

Not necessarily. A lighter AI use policy may be enough. Certification makes sense when AI is central to your products or to decisions about people, or when customers ask for it.

What is the difference between ISO 42001 and ISO 23894?

ISO/IEC 42001 contains requirements for a whole management system and can be certified. ISO/IEC 23894 gives guidance on AI risk management and cannot be certified.