AI Governance Under ISO 42001: Mapping Requirements to M365 Controls
ISO/IEC 42001:2023 is the world's first international standard for AI management systems (AIMS). Published in December 2023, it provides a framework for organisations to manage AI-related risks, establish governance, and demonstrate responsible AI practices. For organisations already operating within the Microsoft 365 ecosystem and deploying AI features like Copilot, ISO 42001 provides a structured path to formalising what should already be good practice.
ISO 42001 Structure Overview
The standard follows the familiar Annex SL structure shared with ISO 27001 and ISO 9001, making it integrable with existing management systems. The core clauses are:
- Clause 4: Context of the organisation (understanding AI-specific risks and stakeholder expectations)
- Clause 5: Leadership (AI governance commitment, policy, roles)
- Clause 6: Planning (AI risk assessment, objectives, treatment)
- Clause 7: Support (resources, competence, awareness, communication, documented information)
- Clause 8: Operation (AI system lifecycle, data management, third-party considerations)
- Clause 9: Performance evaluation (monitoring, internal audit, management review)
- Clause 10: Improvement (nonconformity, corrective action, continual improvement)
The standard also includes Annex A (normative controls), Annex B (AI objectives and risk sources), Annex C (AI-specific guidance), and Annex D (cross-references with other standards).
Mapping Clause Requirements to M365 Controls
Clause 4.1 - Understanding the Organisation and Its Context
Document your AI landscape within M365. This requires a comprehensive inventory of every AI-enabled feature in your tenant:
- Microsoft Copilot for Microsoft 365 (Word, Excel, PowerPoint, Outlook, Teams)
- Copilot Studio agents and custom plugins
- Viva Insights AI-driven analytics
- Defender AI-based threat detection
- Purview AI-powered auto-classification and auto-labelling
- Power Platform AI Builder models
Maintain this inventory in a SharePoint site dedicated to AI governance. Use a SharePoint list with metadata columns for: AI feature name, risk classification, data sources accessed, user groups, deployment date, and review date.
Clause 5.2 - AI Policy
The standard requires a documented AI policy that is communicated across the organisation. Draft this policy addressing:
- Permitted and prohibited uses of AI within the organisation
- Data governance requirements for AI systems
- Human oversight requirements for AI-assisted decisions
- Transparency obligations (internal and external)
- Incident reporting procedures for AI-related issues
Publish the policy via SharePoint and configure a Purview compliance policy requiring all employees to acknowledge the policy annually.
Clause 6.1 - Actions to Address Risks and Opportunities
ISO 42001 requires an AI-specific risk assessment. For M365 AI features, key risk categories include:
- Data exposure: Copilot surfacing sensitive data due to overpermissioned SharePoint sites
- Output accuracy: Copilot generating hallucinated or incorrect information used in business decisions
- Bias and fairness: AI-driven analytics in Viva Insights producing biased workforce assessments
- Prompt injection: Malicious content manipulating Copilot responses
- Regulatory non-compliance: AI usage violating EU AI Act or sector-specific regulations
Document risks in a risk register (SharePoint list or integrated GRC tool) and map each to specific M365 controls.
Clause 8.2 - AI System Life Cycle
The standard requires documented processes for the AI system lifecycle. For M365, this maps to:
- Design and development: Copilot Studio agent creation process, including security review and approval workflow before deployment
- Verification and validation: Testing Copilot agents against known-good test cases before production deployment
- Deployment: Staged rollout via Entra ID security groups, starting with pilot groups
- Monitoring: Continuous monitoring via Purview audit logs and Defender for Cloud Apps
- Retirement: Process for disabling AI features, deleting Copilot Studio agents, and archiving associated data
Clause 8.4 - AI System Data Management
This clause directly maps to Microsoft Purview capabilities:
- Data quality: Deploy Purview Data Lifecycle Management to ensure stale data is archived or deleted before Copilot indexes it
- Data classification: Implement sensitivity labels across the tenant with auto-labelling policies for PII, financial data, and health records
- Data provenance: Use Purview audit logs to track data lineage - who created content, who modified it, when, and how Copilot interacted with it
- Data minimisation: Configure Copilot scope restrictions (Restricted SharePoint Search) to limit AI access to necessary data only
Annex A Control Mapping
A.2 - Policies for AI (A.2.2-A.2.4)
Map to: Microsoft 365 compliance policies published via SharePoint, acknowledged via Purview compliance manager custom assessments.
A.4 - Resources for AI Systems (A.4.3-A.4.6)
Map to: Entra ID role assignments for AI governance (Global Administrator, Compliance Administrator, SharePoint Administrator), documented in the AI governance SharePoint site.
A.5 - Assessing AI System Impact (A.5.2-A.5.4)
Map to: Impact assessments documented per Copilot Studio agent and per AI feature. Use Microsoft Purview Compliance Manager to create custom assessment templates mirroring ISO 42001 control requirements.
A.6 - AI System Lifecycle (A.6.2.2-A.6.2.8)
Map to: Copilot Studio governance workflows, including mandatory approval gates, testing environments, and deployment controls managed through Power Platform environment policies.
A.8 - Data for AI Systems (A.8.2-A.8.5)
Map to: Purview sensitivity labels, DLP policies, auto-labelling, and data lifecycle management policies. This is the strongest mapping between ISO 42001 and M365 native controls.
A.9 - Transparency and Documentation (A.9.2-A.9.4)
Map to: Copilot transparency badges, user notification policies, and Purview audit log retention ensuring all AI interactions are recorded and retrievable.
Evidence Collection for Certification
An ISO 42001 certification auditor will request evidence across all clauses. M365 provides programmatic evidence extraction:
- Purview Compliance Manager: Create a custom assessment for ISO 42001 and map each control to M365 improvement actions. The compliance score provides a quantitative readiness metric.
- Audit log exports: Export Copilot interaction logs, DLP match reports, and sensitivity label application reports as certification evidence.
- Intune compliance reports: If AI features are restricted to compliant devices, export device compliance status as evidence of AI system access controls.
- Entra ID reports: Export Conditional Access policy configurations, PIM activation logs, and access review results demonstrating governance of AI system access.
- Copilot Studio governance: Export agent inventories, approval workflows, and deployment histories from the Power Platform admin centre.
ISO 42001 certification is increasingly expected by enterprise clients and regulatory bodies. For organisations already invested in Microsoft 365, the platform provides native controls that map directly to the standard's requirements, the challenge is not implementing new technology but formalising and documenting the governance around technology you already operate.