ISO/IEC 42001 is the international standard for an AI management system, published in December 2023. It sets requirements for governing AI in an organisation, and an accredited body can certify against it.
What it asks for
The standard follows the structure of other ISO management-system standards, such as ISO/IEC 27001 for information security. An organisation has to:
- Set an AI policy and objectives.
- Assign roles and responsibilities.
- Assess AI risks and the impact of its AI systems.
- Apply controls over the life of each system: data, development, use, third parties.
- Monitor, audit internally and improve.
Annex A lists 38 reference controls. An organisation states which apply to it and why.
Certification
Certification is voluntary and done by an accredited certification body. A certificate lasts three years, with a surveillance audit each year. It says the management system meets the standard. It does not certify that a given model is safe.
How it relates to the others
- The NIST AI RMF is guidance with no certificate.
- SOC 2 is an auditor's report on security controls in general, not on AI.
- The EU AI Act is law. Certification under ISO/IEC 42001 does not by itself show conformity with it.
What it means in practice
Auditors ask for evidence that controls run, not that they are written down. For a deployed model that means records of what the system was allowed to do, what it did and what was stopped: see audit logs.
With White Circle
Control enforces your policies on every interaction and logs each verdict with the policy that fired, which is the kind of evidence an audit asks for.
