October 2026. Microsoft Copilot, ChatGPT and similar assistants are used daily, often without anyone having procured them. To an auditor they are new suppliers and new data flows. How you handle them in your ISMS decides whether you have this under control or a blind spot.
Which controls they touch
An AI assistant touches several controls at once: supplier management (5.19 to 5.23), classification and handling of information (5.12 and 5.13), access control (5.15 and 8.2), data leakage prevention (8.12) and awareness (6.3). No new standard needed; just a deliberate application of what you already have.
Policy: what may and may not go in
Write a short rule based on your classification. For example: public and internal information may go into approved tools, confidential data only into an environment with contractual terms, and personal data or client secrets never into consumer versions. Name the approved tools and who may admit new ones.
Set-up: start with access
Copilot shows users everything they have access to. Messy permissions in SharePoint or Teams suddenly become visible. Run an access review first and trim excessive rights; see also evidence from Microsoft 365. Switch on logging and check the provider’s data use and retention settings.
Supplier assessment of the AI service
Treat the service like any other supplier: where is data processed, is it used to train models, which subprocessors are involved and is there a processor agreement? Record the assessment in your supplier register.
Making unauthorised use visible
People use tools you do not know about. Map that with a short survey and network logs, and include it in your risk assessment; see shadow AI in SMEs. To go further with governance, ISO 42001 offers a framework.
Evidence for the auditor
An approved tool list, the policy, the awareness material, the supplier assessment and an example access review. That is enough to show AI use is part of your controlled environment.