Organizations are beginning to seek certification of their AI management systems against ISO/IEC 42001, and a certificate in a new field raises a question an established field does not. What had to be true of the body that issued it?

ASCB has issued guidance to its accredited certification bodies setting out its expectations for this work. The governing standard for the certification body is ISO/IEC 42006, which does not replace ISO/IEC 17021-1 but sits on top of it, adding requirements specific to AI management systems. ASCB applies those published standards as they stand and does not require certification bodies to meet additional mandatory documents of its own.

The guidance, ASL(G)102.101, is issued to certification bodies accredited by ASCB and to those applying for ISO/IEC 42001 accreditation. It is not published publicly, and accredited bodies and applicants should request it from ASCB. An overview of ISO/IEC 17021-1, the base standard these requirements sit on top of, is in the Resource Center Standards Library.

Three of the differences are substantial enough to matter to a buyer.

Audit time is calculated differently. For most schemes the starting point is the number of people in the organization, sometimes with turnover as a secondary factor. For an AI management system the starting point is the number of people involved in the AI life cycle, and turnover does not feature at all. That figure is then adjusted for the number of AI systems, how many of them are high-risk or serve sensitive purposes, the regulatory frameworks and third-party agreements in scope, and the number of controls in the statement of applicability. The result is treated as a minimum rather than a schedule, and the reasoning has to be recorded.

The impartiality bar is wider. A certification body is already barred from providing management system consultancy or internal audits to a client it certifies. For an AI management system client that bar extends to consulting on artificial intelligence, information security, data protection and risk management, including acting as the client's external data protection officer. The audit itself cannot be outsourced.

Access has to be settled first. AI work frequently involves confidential material, including source code and raw data. Before the audit, the certification body has to ask the client to identify anything that cannot be made available and then determine whether the management system can still be adequately audited. Where it cannot, the audit does not proceed until access arrangements are in place.

For anyone relying on one of these certificates, the practical questions are the ordinary ones. Is the certification body accredited for this activity, does the scope of that accreditation cover AI management system certification specifically rather than management systems in general, and is it current. Each of those is a matter of record rather than judgment, and the Resource Center explains how to check an ASCB-accredited certificate and its scope.