Skip to content

Control 1.10: Vendor Risk Management for Microsoft AI Services — Portal Walkthrough

Step-by-step procedures for conducting vendor risk assessment of Microsoft AI services used by Microsoft 365 Copilot.

Prerequisites

  • Vendor Risk Management team access and authority
  • Microsoft Service Trust Portal access (servicetrust.microsoft.com)
  • Third-party risk management framework or questionnaire (e.g., SIG, CAIQ)
  • Governance committee sponsorship for vendor assessment

Steps

Step 1: Access Microsoft Service Trust Portal

Portal: Microsoft Service Trust Portal Path: servicetrust.microsoft.com > Sign in with organizational credentials

Access the Service Trust Portal to review Microsoft's compliance documentation, audit reports, and data protection assessments relevant to AI services. Download the following key documents: - SOC 2 Type II report for Microsoft 365 - ISO 27001 certificate for Azure and Microsoft 365 - Microsoft AI Transparency Notes, Data Protection Addendum (DPA), and Service Trust Portal documentation - Data Processing Agreement (DPA) and product terms

Step 2: Review Microsoft's AI Transparency Documentation

Portal: Microsoft Trust Center Path: microsoft.com/trust-center > AI > Responsible AI

Review Microsoft's responsible AI commitments, transparency reports, and AI governance frameworks. Document: - Microsoft's AI principles and how they apply to Copilot - Data handling practices for Copilot prompts and responses - Training data usage policies (tenant data is not used for model training) - Sub-processor and third-party data processing disclosures

Step 3: Complete Vendor Risk Questionnaire

Using your organization's vendor risk assessment framework, complete the assessment for Microsoft AI services. Key areas to evaluate: - Data security: How is data protected during AI processing? - Data residency: Where is data processed and stored? - Model governance: How are AI models updated and tested? - Incident response: What is Microsoft's AI incident response process? - Contractual protections: What SLAs, indemnification, and liability terms apply?

Step 4: Review AI Provider Settings and Terms

Portal: Microsoft 365 Admin Center Path: Copilot > Settings > View all

Review the provider paths separately and retain screenshots or exports showing the current setting and assigned users or groups. Use Microsoft Learn's current role names in evidence: Anthropic provider changes require AI Administrator or Global Administrator, SpaceXAI subprocessor changes require AI Administrator or Global Administrator, while SpaceXAI independent-processor changes require Global Administrator.

  1. Open AI providers operating as Microsoft subprocessors and review Anthropic.
  2. Record whether standard Anthropic models are enabled and who can access them.
  3. Check separately for Anthropic models with Data Retention. These models require explicit opt-in and separate Anthropic terms because they operate outside the Microsoft subprocessor arrangement.
  4. Open AI providers for other large language models and review SpaceXAI as an independent processor (used with Copilot Studio and Copilot Cowork).
  5. Confirm the legal terms were reviewed before enablement.
  6. Record the users or Entra security groups authorized to use the provider.
  7. Back under AI providers operating as Microsoft subprocessors, check separately whether SpaceXAI is enabled as a subprocessor (preview; Frontier-program customers outside the EU, EFTA, and UK or other environments Learn lists as excluded). It is a distinct setting from the independent-processor setting, is disabled by default, and is not carried over from earlier SpaceXAI settings; record the assigned users or groups and the documented exclusions (including in-country processing).
  8. For Copilot Studio use, confirm the corresponding external-model control is also enabled only in approved Power Platform environments.
  9. Reconcile each enabled provider and model category to the vendor inventory, data-flow map, and risk-acceptance record.

Organizations should verify current terms and default states in Microsoft Learn before each approval because provider availability and model categories change over time.

Step 5: Document Risk Findings and Mitigations

Portal: Internal GRC platform or risk register

Record all identified risks from the vendor assessment along with: - Risk severity rating (Critical, High, Medium, Low) - Existing mitigations provided by Microsoft - Additional mitigations required by the organization - Residual risk acceptance decisions with governance committee approval

Step 6: Establish Ongoing Monitoring

Set up ongoing vendor monitoring processes: - Subscribe to Microsoft 365 Message Center for service changes - Monitor Microsoft Security Response Center for AI-related advisories - Schedule annual vendor risk reassessment - Track Microsoft compliance certification renewal dates

FSI Recommendations

Tier Recommendation
Baseline Complete vendor risk assessment and document findings before Copilot deployment
Recommended Integrate Microsoft AI services into ongoing vendor monitoring program with semi-annual reassessment
Regulated Comprehensive third-party risk assessment per OCC/FFIEC guidelines; board-level reporting on AI vendor risk; contractual review by legal counsel

Next Steps