Who Audits the AI Auditors? What ISO/IEC 42006 Changes for 42001 Certificates

An ISO/IEC 42001 certificate is quickly becoming the default answer to a procurement question: how do we know this vendor governs its AI responsibly? The certificate is only as meaningful as the audit behind it, and until recently there was no AI-specific standard governing the people who conduct that audit. ISO/IEC 42006, published in July 2025, is meant to close that gap.

What ISO/IEC 42006 is

ISO/IEC 42001 is the management system standard an organization is certified against. ISO/IEC 42006 is different: it sets requirements for the certification bodies that audit and certify AI management systems. It supplements ISO/IEC 17021-1, the general standard for management system certifiers, with AI-specific rules. If 42001 defines what a good AI management system looks like, 42006 defines what a credible auditor of one looks like.

Three requirements worth knowing

Competence. Audit teams must collectively cover AI, management systems, auditing, AI governance, and the technical activities inside the certification scope. A team that can audit an information security management system is not automatically able to judge training-data provenance, model monitoring, or impact assessment.

Audit time. The standard introduces a structured method for calculating audit duration, drawing on factors such as applicable regulation, the number of AI systems, high-risk applications and third-party relationships. It also states that planning and report writing should not normally reduce client-facing audit time below 70% of the calculated figure, and recertification audits should take at least two-thirds of the initial certification time.

No automatic accreditation. The standard does not make every 42001 certificate an accredited one. Accreditation bodies assess certifiers separately, so two certificates with the same standard number can rest on very different levels of assurance.

Why this matters for buyers and boards

Certificates tend to be treated as binary evidence: present or absent. That habit is risky here. A scope statement that covers a single internal chatbot says little about the model embedded in your customer workflow. A certificate issued by a body with no AI competence says little about anything. The assurance value sits in three details that rarely appear in a vendor questionnaire: what the scope includes, who issued the certificate, and whether that issuer is accredited for this standard.

Practical steps

If you are procuring AI services, ask for the certificate, the scope statement and the issuing body's accreditation status, and check that the certified scope covers the system you actually depend on. If you are pursuing certification yourself, ask prospective certifiers how they build audit teams and calculate audit time. An unusually short, cheap audit is a warning sign, not a bargain.

If you sit on a board or risk committee, treat 42001 certification as one input to vendor assurance, not a substitute for it. It tells you a management system exists and was examined. It does not tell you the models are safe, fair or compliant with a specific law, and it does not replace your own testing, contractual controls and monitoring.

The governance reading

Assurance ecosystems mature in layers: a standard for the practice, a standard for the assessors, and accreditation for the assessors' assessors. ISO/IEC 42006 supplies the middle layer for AI. It will not eliminate weak audits, but it gives buyers and regulators a benchmark to hold certifiers to. The question to start asking is no longer only whether a vendor is certified. It is who certified them, against what scope, and with what competence.