There is a lot of talk about AI management (ISO/IEC 42001) right now. Its less glamorous but older sibling often gets overlooked: ISO/IEC 27001, the international standard for information security. Yet it is the basis on which AI becomes accountable in the first place — because an AI is only as trustworthy as the data and systems it runs on.
What ISO 27001 is
ISO/IEC 27001 describes an Information Security Management System (ISMS) — a management system with which an organisation systematically steers its information security: what must be protected, against which risks, with which measures, and who is responsible?
The approach matters: the standard does not prescribe a rigid catalogue. It requires a risk-based approach — you identify your assets and threats and select the measures that fit your actual risk. It follows the same base structure (clauses 4–10) as ISO 9001 and ISO 42001. Whoever runs one of these management systems can dock the ISMS onto existing processes.
What the 2022 edition changed
The current version is ISO/IEC 27001:2022. The most important change is in the set of controls (Annex A): the former 114 controls became 93 — consolidated, streamlined and reorganised into four themes:
- Organizational (37 controls) — policies, roles, supplier security, incident handling, use of cloud services.
- People (8 controls) — awareness, training, responsibilities of staff.
- Physical (14 controls) — access, equipment, monitoring of premises and technology.
- Technological (34 controls) — access control, cryptography, logging, secure development.
11 new controls were added to reflect modern topics — among them threat intelligence, information security for cloud services, configuration management, information deletion, data masking, data leakage prevention and secure coding. Precisely the topics that also matter when you deploy AI.
Select and justify: the Statement of Applicability
At the heart of the ISMS is the Statement of Applicability (SoA) — the declaration of which controls apply to you and why. Not every measure fits every company. The standard does not demand completeness at any price but a traceable justification: what is relevant, what is not — and why. That is the difference between lived security and a binder for the auditor.
How certification works
Whoever pursues the certificate goes through an audit by an accredited body: Stage 1 examines the documentation, Stage 2 the actual implementation. The certificate is usually valid for three years, with annual surveillance audits, followed by recertification. As with ISO 42001: whether you certify is decided by market requirements — tenders, large customers, regulated industries. The substance is worth it even without the certificate.
Why 27001 is the foundation for AI
This is where the circle closes back to artificial intelligence. An AI system processes your most sensitive data: engineering knowledge, customer data, trade secrets. Without solid information security, every AI adoption becomes a risk — no matter how good the model is.
That is why ISO/IEC 27001 and ISO/IEC 42001 belong together for us: 27001 secures the data and systems, 42001 governs the responsible use of AI on top of them. Our principle of “self-hosted first” is lived 27001 — cryptography, access control and data sovereignty, so that sensitive data stays where you keep control. Cloud where it is uncritical; your own infrastructure where it counts.
We work by both standards ourselves and guide the build at clients — as a TÜV Süd certified AI Officer; the ISO/IEC 42001 certification is in preparation. The first step is always an honest inventory: know where you stand.