Integration
Extending ISO 27001 to cover AI, one clause at a time.
ISO publishes its management system standards on a common harmonized structure, so ISO/IEC 27001 and ISO/IEC 42001 number and title their clauses 4 to 10 the same way. That shared frame is the reason a company with an information security management system (ISMS) can grow it into an AI management system (AIMS) instead of standing up a rival. The work sits in the gaps: what each clause must now say about AI, and which records have no security ancestor at all.
Back to ISO 42001 services01 · Clause delta
The delta, clause by clause
Left column: what a mature ISMS usually holds. Right column: what AI adds.
| Clause | Usually in place from ISO 27001 | What AI adds |
|---|---|---|
| 4 Context | Interested parties, scope boundaries, interfaces | Your role for each AI system, its intended purpose, and people affected by outputs who never signed a contract |
| 5 Leadership | A security policy and assigned responsibilities | A separate AI policy approved by top management, and an owner named for every AI system |
| 6 Planning | Risk assessment and treatment, a Statement of Applicability, security objectives | The AI system impact assessment (6.1.4), risk sources peculiar to machine learning, AI objectives, and an SoA against the 42001 Annex A |
| 7 Support | Competence records, awareness, document control | Data science and model skills on the competence matrix; AI awareness for staff who use tools |
| 8 Operation | Operational planning, change management | Life-cycle stages for each system: design, data preparation, testing, release, retirement |
| 9 Performance evaluation | Monitoring, an internal audit programme, management review | Model quality and drift measures, auditors who understand AI, AI items on the review agenda |
| 10 Improvement | Nonconformity, corrective action, continual improvement | Failures caused by model behaviour, not just by breaches, handled through the same route |
Most rows are additions to documents you already maintain. The exceptions are clause 6, where AI brings an assessment type ISO 27001 never asked for, and clause 8, where a model's life cycle needs stages a security process does not track.
02 · Brand-new records
Records with no ISMS ancestor
Where the build effort really goes.
Start from a blank page
- An AI system inventory: purpose, role, owner and status for each system
- Impact assessments looking outward at individuals and society, with ISO/IEC 42005 as guidance
- Data provenance records: where training and test data came from and the right to use it
- Release evidence for each model version: what was tested, against which threshold
- Proof that people can see they face AI and that a human can step in
Stretch something you run
- Supplier reviews, widened to model vendors and data suppliers
- Logging, access and change controls over the infrastructure models run on
- Incident handling, with model misbehaviour added as a category
- Risk scales and the risk register, with AI risk sources added
- Training records, with AI topics added for the right roles
If the left column feels long, that is normal. It is the part an AIMS auditor samples most closely.
03 · Two SoAs
Two Statements of Applicability, one evidence trail
The control lists stay separate. The proof does not need to.
ISO 27001 has 93 reference controls in four themes; ISO 42001 has 38 across nine objectives. Neither list counts toward the other, so each standard keeps its own Statement of Applicability (SoA). What you avoid is duplicate evidence.
Illustrative: one log, cited twice
A fintech in Pune runs a credit-scoring model. Its ISMS already keeps tamper-resistant logs of who changed production systems. The AIMS SoA points to the same log as part of the evidence that model releases are controlled, and adds one field: the model version. One record, two citations, and the internal auditor tests it once.
04 · Audit logistics
Planning the audits together
One body, one calendar, if the body is accredited for both.
Certification bodies can run combined audits of an integrated system, and the International Accreditation Forum (IAF) publishes a mandatory document, IAF MD 11, on how audit time works when systems are integrated. The saving depends on how integrated your system really is: shared policies, one internal audit programme, one management review. Before you sign, confirm the body is accredited for AIMS as well as ISMS work.
- Line up certificate cycles so surveillance visits land in the same window.
- Run one internal audit programme with AI-competent auditors for the AIMS clauses.
- Hold one management review with a separate AI section in the minutes.
05 · Shortcuts
Integration shortcuts that backfire
Each looks efficient on paper and fails in the audit room.
- Renaming security controls as AI controls. Annex A of 42001 covers bias, data rights and oversight. A firewall rule says nothing about those.
- One line on AI in the security policy. The AIMS needs a policy leadership can be held to on AI itself. A sentence inside another document rarely carries that weight.
- Same scope by default. The AIMS scope should follow where AI is built and used, which may be narrower or wider than the ISMS boundary.
- Skipping the impact assessment. A security risk assessment looks at harm to the company. The AIMS also wants harm to people outside it.
Sources: ISO/IEC 27001:2022 and ISO/IEC 42001:2023 public clause titles and Annex A structure; ISO/IEC Directives Part 1, Consolidated ISO Supplement, Annex SL (harmonized structure); ISO/IEC 42005:2025 public summary; IAF MD 11, application of ISO/IEC 17021-1 to audits of integrated management systems.
Questions about combining the two
Must the AIMS scope match our ISO 27001 scope?
No. Draw each around what it governs.
Your ISMS might cover the whole company while AI lives in one product team, or the reverse. Overlap helps, but the AIMS boundary follows your AI systems, and each certificate states its own scope.
Can our current internal auditor cover the AIMS too?
Yes, if their competence covers AI.
Clause 9.2 wants auditors who are objective and competent for what they audit. Pair your ISMS auditor with someone who understands model testing and data, or train them, and record that competence.
Where does ISO/IEC 27701 fit if we hold it?
It stays the privacy extension of your ISMS.
Privacy records from 27701 feed the AIMS where personal data trains or flows through a model. They do not replace the AI impact assessment, which asks wider questions than privacy.
Which document should we write first?
The AI system inventory.
Scope, roles, impact assessments and the second SoA all hang off the inventory. Without it, every other AIMS record has nothing to point at.
Grow the system you already run.
We map the clause-by-clause gap between your ISO 27001 system and an ISO 42001 AIMS, and help you build only the missing pieces.