AI policy ISO 27001
If you have ISO 27001 certification, or you are working towards it, AI use should not sit outside your information security management system. Staff need clear rules for approved AI services, confidential information, access control, human review and accountability.
ISO/IEC 27001 is built around managing information security risk. AI tools can affect that risk because staff may use them to process confidential information, summarise customer material, generate code, draft documents, analyse data or support decisions.
The starting point can be simple: AI use should be visible, approved and consistent with the way the organisation already manages information security.
This does not mean every AI use needs a separate control framework. It means AI should be handled through the same disciplined approach used for other information security risks: clear ownership, documented expectations, approved services, risk treatment and periodic review.
AI services can create risk when people paste sensitive information into unapproved tools, rely on inaccurate outputs, expose information through weak permissions, or use generated content without review. These risks sit naturally alongside existing information security concerns such as confidentiality, integrity, availability, supplier assurance, access control and incident management.
A practical AI policy gives staff enough guidance to make safer choices and gives auditors, customers and leadership a clearer view of how AI use is controlled.
Common examples include staff using AI to summarise confidential documents, draft customer responses, generate code, review logs, analyse spreadsheets or prepare management reports. Each use may be low risk in isolation, but unmanaged use across a business can create a much wider control gap.
The policy should connect everyday staff behaviour to the wider information security management system. It should be clear enough for users, but also specific enough to support risk owners, auditors and customers who want to understand how AI use is governed.
ISO 27001 works best when controls reflect the way the organisation actually operates. AI should be handled in the same spirit: understand where it touches information, decide what is acceptable, document the controls and review whether those controls are working.
Staff paste confidential documents, support tickets, customer material or source code into a tool that has not been assessed or approved.
AI-assisted analysis, code, summaries or recommendations are used without checking accuracy, assumptions or source material.
An integrated AI tool can summarise information from systems where permissions, labels or sharing settings are too broad.
A new AI service is adopted without checking terms, data handling, retention, security controls or where organisational information may be processed.
The Statement of Applicability does not usually need a separate entry for every AI use case. It should, however, make sense when AI use is reviewed against the controls already selected for the ISMS.
AI can touch several familiar control areas, including acceptable use, access control, information classification, supplier relationships, event logging, monitoring, secure development, incident management and awareness training. The useful question is not "which clause mentions AI?", but "which existing controls are affected by the way we use AI?".
| ISO 27001 area | AI policy connection |
|---|---|
| Acceptable use | Explain which AI tools staff may use and what information must stay out of them. |
| Access control | Review permissions before connected AI tools can search, summarise or expose internal information. |
| Supplier management | Assess AI providers, data handling, retention, sub-processors and contractual terms. |
| Information classification | Map classified, confidential or personal data to clear AI usage restrictions. |
| Incident management | Tell staff what to do if sensitive information is entered into the wrong AI tool. |
For ISO 27001-minded organisations, one of the most useful AI controls is a simple approved services list. Staff should know which AI tools they can use, whether those tools are approved for confidential information, and who can approve new use cases.
This is especially important where AI tools connect to email, documents, tickets, chat systems, code repositories or customer records. The policy should match real usage rather than pretending AI is not already part of daily work.
A simple approved services list might record the tool name, business owner, permitted uses, data restrictions, approval date, next review date and whether the service can be used with confidential information. That list can then support onboarding, staff guidance and internal audit activity.
Public AI tools and enterprise AI tools can create different risk profiles. A public chatbot may need stricter rules around sensitive information, while Microsoft Copilot or another integrated platform may need stronger attention to permissions, labels, retention and who can access what.
The policy should make those distinctions plain. Staff should not have to guess whether a tool is approved, what information can be used, or whether an output needs checking before it becomes part of business work.
If Microsoft Copilot is being considered, the policy should sit alongside permission reviews, information classification, data retention, sensitivity labels and user training. If public tools such as ChatGPT, Gemini or Claude are allowed, the policy should be explicit about what information can and cannot be entered.
AI services should be reviewed before they become part of normal work, particularly where they process organisational information, client data, personal data, security information or source code. A lightweight supplier review is often enough for low-risk tools, but the decision should be visible and repeatable.
AI can support work, but it should not remove accountability. A good policy makes clear that people remain responsible for checking accuracy, suitability, confidentiality, copyright and fairness before AI-assisted work is used.
It should also connect AI use back to access control. If an AI system can summarise or generate content from internal data, the organisation needs confidence that permissions, data classification and information handling rules are working as intended.
Internal audit does not need to make AI governance complicated. A proportionate review can check whether the organisation understands how AI is being used and whether staff practice matches the policy.
Useful audit questions include: are approved tools listed and reviewed, are staff aware of the rules, are exceptions recorded, are high-value use cases risk assessed, are suppliers reviewed, and is there a clear route for reporting concerns or accidental disclosure?
An AI policy becomes more useful when it is supported by a small amount of evidence. This does not need to be excessive, but it should show that AI use is understood, reviewed and connected to existing controls.
FAQ
Not necessarily. ISO/IEC 27001 is about managing information security risk through an ISMS. If AI tools affect that risk, the organisation should be able to show how AI use is governed through suitable policies, controls and review.
Yes, where AI tools handle organisational information, influence important work, connect to business systems or create meaningful confidentiality, integrity or availability risks.
It should cover approved AI services, information that must not be entered into AI tools, access control, human review, records, responsibilities, incident reporting and review dates.
It is a plain list of AI tools the organisation permits, what they may be used for, what information can be entered, and who owns approval or review.
The Statement of Applicability does not need to list every AI use case, but relevant controls should reflect how AI affects access control, supplier management, information classification, acceptable use, logging, monitoring and incident response.
Useful evidence may include an AI usage policy, approved AI services list, risk assessment entries, staff guidance, training records, incident routes, review dates and examples of human review controls.
Microsoft Copilot often needs attention to Microsoft 365 permissions, information labels, retention, oversharing, user training and monitoring because it works across existing business information.
They can be allowed, but the rules should be clear. Public AI tools usually need stricter limits on confidential information, personal data, client data, credentials, source code and commercially sensitive material.
Supplier review should consider the tool purpose, information processed, terms, retention, security controls, sub-processors, access arrangements, audit evidence and exit options.
Internal audit can look for an approved AI services list, risk treatment evidence, staff awareness, exception handling, incident routes, review dates and whether real AI use matches the documented policy.
A policy is a useful starting point, but it should be supported by ownership, approved tools, risk assessment, supplier review, training, monitoring and periodic review.