ISO 42001: the management system standard for AI

What ISO 42001 covers that ISO 27001 does not

ISO 27001 protects information. ISO 42001 governs how AI systems are chosen, built, deployed, monitored and retired, which brings in risks a security standard was not written for. The certificate covers the management system around your AI rather than any one model, and the standard applies whether you train models yourself or build features on someone else’s.

Bias and fairness

Whether outcomes differ unfairly across the people a system affects.

Transparency

Whether people know AI is involved and can understand its role.

Data provenance

Where training and input data came from, and on what basis it is used.

Drift and lifecycle

How performance is monitored, and how systems are changed or retired.

Human oversight

Who can review, override or stop an AI system’s output.

Third-party AI

Governing models and features you buy rather than build.

Who is being asked for it

Enterprise security questionnaires now include AI sections on training data, bias testing and human oversight that a SOC 2 report alone cannot answer, and procurement teams increasingly name ISO 42001 as the evidence they would accept.

• Vendors selling AI features into regulated industries such as finance, health and HR
• Companies in scope of the EU AI Act that want a demonstrable governance baseline
• Suppliers to enterprises with published AI procurement standards
• Teams embedding third-party foundation models in customer-facing products

What an AI management system contains

ISO 42001 follows the same clause structure as ISO 27001: context, leadership, planning, support, operation, performance evaluation and improvement. The subject is different. You keep an inventory of AI systems, assess AI-specific risks and impacts, and apply the 38 Annex A controls for AI, which cover policy, data, the system lifecycle, third parties and accountability.

AI system inventory

Every system with its owner, purpose, data, affected people and risk tier.

AI impact assessment (6.1.4)

The effects a system could have on individuals, groups and society.

Lifecycle controls

How systems are designed, tested, released, monitored and retired.

Accountability

Named owners, an AI policy, and records of the decisions made.

Start with the inventory

The hardest first step is knowing what AI you actually run. Most organisations underestimate it, because AI arrives through product features, vendor updates and experiments as often as through planned projects.

• Third-party models called through an API inside product features
• AI embedded in SaaS tools your teams already use
• Internal experiments that quietly reached production
• Vendor features switched on by default after an update

For each system, record the owner, the purpose, the data it uses, who is affected by its output and a risk tier. Our guide on how to build an AI system inventory walks through it field by field.

AI risk and impact assessments

ISO 42001 asks for two related assessments. The AI risk assessment looks at risks to your organisation and its objectives. The AI system impact assessment looks outwards, at the effects a system could have on individuals, groups and society. Systems with a high risk tier need the most depth.

Record AI risks in the same register as your other risks, with the same scoring, owners, treatments and review dates, so leadership sees one picture.

ISO 42001, the EU AI Act and NIST AI RMF

ISO 42001 certification does not by itself prove EU AI Act compliance. An ISO 42001 management system does produce much of the governance evidence the Act expects, such as risk management, documentation, human oversight and record-keeping, so teams preparing for the Act often use the standard to organise that work.

On timing: the Act’s prohibited practices applied from February 2025 and general-purpose AI obligations from August 2025. Transparency duties apply from August 2026. High-risk obligations apply from December 2027 for Annex III systems and August 2028 for AI in regulated products, after the 2026 Digital Omnibus moved them back.

ISO 42001 and the EU AI Act are both in Certifyi’s framework catalogue. You can map a control once and reuse the mapping across frameworks.

Building on ISO 27001

If you hold ISO 27001, a large part of an ISO 42001 management system already exists: the governance clauses, risk method, document control, internal audit and management review. What you add is the AI inventory, AI-specific risk and impact assessments, and the Annex A controls for AI.

Many certification bodies run integrated audits, so an ISO 27001 surveillance visit and ISO 42001 Stage 1 can happen together, with shared controls examined once. You do not need ISO 27001 first, but without it you build the management system as well as the AI-specific parts.