Why is integrating governance, process and technology under ISO 42001 worth more than a stack of AI policies?
For many companies the first AI governance document is an acceptable-use policy. The real exposure sits elsewhere: generative AI services wired in by individual teams, third-party models that change version without notice, training data whose provenance nobody can explain. ISO/IEC 42001 asks for an AI management system that runs on its own, not a policy:
- Governance: determine risk level and provider or deployer role under the EU AI Act, then build the AI management system, impact assessment and human-oversight procedures under ISO 42001.
- Process: impact assessment sits in your product development gates; bought-in models go through existing procurement and supplier evaluation.
- Technology: data governance, model testing, and bias and robustness validation produce records you can retrieve.
We use three documents (an AI system inventory, an AI risk and impact assessment, and an EU AI Act obligations matrix) so that ISO 42001 and the AI Act share one body of evidence. Organisations that already hold ISO 27001 can reuse document control, internal audit and management review, and usually reach ISO 42001 certification in 4 to 6 months.
For the control-by-control build, see ISO 42001 technical integration. This page is about keeping the system growing with your AI use.
How can Ahura keep ISO 42001 evidence integrated with ISO 27001 and the EU AI Act?
AI management system evidence is scattered: model test reports in R&D, supplier assessments in procurement, impact assessments in legal. Every new demand, whether a customer AI questionnaire, the EU AI Act or ISO 27001 change management, means repackaging the same report.
Ahura, the enterprise risk management platform developed by Winners Consulting, is currently at proof-of-concept (POC) stage. What it already provides:
- 100% on-premises deployment; models, training data and assessment reports are core assets and never leave your environment;
- an end-to-end evidence chain in which every sign-off is recorded, compatible with EU eIDAS certificates;
- a distributed group architecture with EU-language interfaces;
- a modular design for adding EU regulations and international standards.
The synergy starts with one piece of evidence and one sign-off chain, referenced by several standards and laws. A signed model deployment approval can serve as an ISO 42001 record, input to AI Act technical documentation and support for ISO 27001 change management.
To be precise: cross-standard comparison, gap analysis from uploaded documents, regulatory version timelines and regulatory change tracking are in development and will open as the POC progresses, and there is no ISO 42001-specific module today. The goal, once delivered, is to map regulatory versions and changing obligations directly against the evidence you hold.
Can ISO 42001 internal audits be outsourced objectively, and produce an AI governance roadmap?
Clause 9.2 of ISO/IEC 42001:2023 covers internal audit and requires an internal audit programme (9.2.2); auditors must be selected and audits conducted so that the audit process is objective and impartial. ISO 19011 treats independence and objectivity as audit principles. AI makes this harder: the people who understand the models usually helped build them, and the people who understand management systems may not read a model test report. Few organisations can staff an internal AI audit that is both competent and uninvolved, which is why AI governance internal audit outsourcing makes sense.
Our principle: the consultant who designed a process never audits that design. Our consultants hold ISO Lead Auditor qualifications and audit documents and system records together: does the inventory include newly launched AI applications, is the impact assessment redone when systems change, is every deployment approval backed by test evidence?
Afterwards we sort findings with you into nonconformities to correct, observations worth improving, and directions for next year, such as bringing team-level generative AI use into the inventory, triggering reassessment when a supplier changes model version, or rolling AI risk into the enterprise risk map. That roadmap goes into management review (9.3), where leadership sets the order.
An internal audit is a first-party audit, not certification. Certification is carried out by a certification body; Winners Consulting provides consulting only and does not issue certificates.
How does an outsourced audit of AI model, data and cloud AI suppliers keep your AI supply chain trustworthy?
A third-party model does not shift accountability just because someone else trained it. ISO/IEC 42001 Annex A covers third-party and customer relationships in A.10, including A.10.2 (allocating responsibilities) and A.10.3 (suppliers), which require you to define responsibilities between you and your suppliers and to manage what they deliver.
When you commission us to audit AI suppliers, that is a second-party audit in ISO 19011 terms. It is an AI vendor audit, not certification, and no certificate is issued: we confirm on your behalf that suppliers are meeting their contractual commitments.
| Supplier type | Audit focus (ISO/IEC 42001 Annex A) | Evidence retained |
|---|---|---|
| Foundation model and generative AI API providers | A.10.3 suppliers; A.10.2 responsibilities: data-use limits, version change notice | Responsibility matrix, change notification records |
| Training data and labelling providers | A.7.5 data provenance; A.7.4 data quality | Provenance records, sampled quality checks |
| Cloud AI platform and MLOps providers | A.6.2.6 operation and monitoring: are logs retrievable? | Monitoring and log retention records |
| Outsourced AI developers and integrators | A.10.2 responsibilities; A.8.4 communication of incidents | Contract clauses, incident notification test records |
Results feed back into your AI system inventory and supplier evaluation, so every AI supplier is measured against the same yardstick. To scope an AI supplier audit programme, contact us.
Official sources
- DIN Media — ISO/IEC 42001:2023 table of contents: 9.2 Internal audit, 9.2.1 General, 9.2.2 Internal audit programme; Annex A reference control objectives and controls; Annex B implementation guidance
- NIST AI Resource Center — NIST AI RMF to ISO/IEC 42001 Crosswalk (B.10.2 Allocating responsibilities, B.10.3 Suppliers, B.7.4 Quality of data, B.7.5 Data provenance, B.6.2.6 Operation and monitoring, B.8.4 Communication of incidents, 9.2.2, 9.3)
- NIST AI Resource Center — Crosswalk documents index
- ISO/TC 176 & IAF — ISO 9001 Auditing Practices Group, Guidance on: Internal Audits (objectivity and impartiality of the internal audit process)
- ISO 9001 Auditing Practices Group — Guidance on: Impartiality (self-review threats)
- ISO — ISO 19011:2026 Guidelines for auditing management systems (edition 4, published 2026-05; first-, second- and third-party audits)
- CQI | IRCA — Revision of ISO 19011: what you need to know (principles; objectivity where independence is not practicable; second-party audit text in Annex A.12)
