# yaml-language-server: $schema=../../../../schema/control.schema.json
id: CB-M365-DATA-009
title: Detect risky AI usage with Insider Risk Management
finding:
  title: Risky use of AI assistants by insiders goes undetected
  description: Users who feed sensitive data into Copilot or third-party generative AI sites, or try prompt injection, raise no signal for the security team.
status: active
profiles: [hardened]
risk:
  severity: low
assessment: manual
tags: [insider-risk, purview, copilot, ai, monitoring]

applicability:
  requires:
    - "Microsoft Purview E5 Compliance"

description: |
  A Microsoft Purview Insider Risk Management policy built from the **Risky AI usage** template is
  active and scoped to all users (or to the populations defined with the works council). It
  detects risky prompts and responses in Microsoft 365 Copilot and other AI apps, such as sensitive
  information in prompts, access to protected content through AI, and prompt injection attempts.
  Pseudonymization of users in alerts stays enabled.
rationale: |
  Generative AI creates new paths for data leakage that DLP rules on mail and files do not see. The
  Risky AI usage template correlates AI interactions with other insider risk indicators so the team
  can investigate patterns rather than single events.
impact: |
  Insider risk monitoring processes personal data about employees. In France, deploy it only after a
  data protection impact assessment, information of employees and consultation of the works council
  (CSE). Limit access to alerts to a small, named team.
notes: |
  Collecting AI interactions outside Microsoft 365 (third-party AI sites) needs device onboarding
  and the browser extension or network integration described by Microsoft.

audit:
  manual: |
    1. Sign in to the Microsoft Purview portal (https://purview.microsoft.com) as a member of the
       Insider Risk Management Analysts or Insider Risk Management Auditors role group (read-only
       roles for this solution).
    2. Browse to **Solutions** > **Insider Risk Management** > **Policies**.
    3. Check that a policy based on the **Risky AI usage** template exists, its status is active
       (not paused or in test), and its scope covers the intended users.
    4. Browse to **Insider Risk Management** > **Settings** > **Privacy** and check that
       pseudonymized user names are shown.
    5. Fail if no active Risky AI usage policy exists.

remediation:
  manual: |
    1. Complete the data protection impact assessment and works council consultation.
    2. Sign in to the Microsoft Purview portal as an Insider Risk Management Admin.
    3. Configure **Insider Risk Management** > **Settings** (privacy, policy indicators for Microsoft
       365 Copilot and AI apps).
    4. Browse to **Policies** > **Create policy**, select the **Risky AI usage** template, set the
       users in scope, the indicators and thresholds, and create the policy.
    5. Assign alert triage to the Insider Risk Management Investigators role group.
  rollback: |
    Pause or delete the policy under **Insider Risk Management** > **Policies**.

references:
  - title: Insider risk management policy templates
    url: https://learn.microsoft.com/purview/insider-risk-management-policy-templates
    publisher: Microsoft
    kind: vendor-doc
    accessed: 2026-10-24
  - title: Microsoft Purview data security and compliance protections for generative AI apps
    url: https://learn.microsoft.com/purview/ai-microsoft-purview
    publisher: Microsoft
    kind: vendor-doc
    accessed: 2026-10-24
  - title: Maester MT.1174
    url: https://maester.dev/docs/tests/
    publisher: Maester
    kind: tool
    accessed: 2026-10-24
mappings:
  - framework: nist-800-53r5
    framework_version: "5.1.1"
    id: PM-12
    relation: related
  - framework: nist-800-53r5
    framework_version: "5.1.1"
    id: SI-4
    relation: related
related_controls:
  - id: CB-M365-DATA-004
    type: related

authors:
  - name: Alice
    github: Sn0wAlice
introduced_in: 0.1.0
created: 2026-10-24
updated: 2026-10-24