ExamFinch.com My Study

Study guide

2.3 Recognizing & Mitigating Risks to Sensitive Data

Practice this exam

Microsoft AI Business Professional AB-730

Track your progress as you read

Finished sections get a check mark.

1 Introduction & Exam Overview

0.1 About the AB-730 Exam 0.2 How to Use This Study Guide

2 Chapter 1: Understand Generative AI Capabilities Across Microsoft 365

1.1 How Copilot Works: LLMs, Microsoft Graph & the Security Boundary 1.2 Context & Grounding: Work Files, Web Data and App Context 1.3 Chat Experiences vs. Agent Experiences 1.4 When to Create Your Own Agent 1.5 Copilot Capabilities Across Microsoft 365 Apps

3 Chapter 2: Responsible AI & Data Protection Practices

2.1 Generative AI Risks: Fabrications, Prompt Injection & Over-Reliance 2.2 Verification Steps: Citation Checks & Human Review 2.3 Recognizing & Mitigating Risks to Sensitive Data 2.4 How Data Protection Restricts Prompt Results

4 Chapter 3: Creating and Managing Prompts

3.1 Creating an Effective Prompt: The GCSE Framework 3.2 Selecting Resources to Reference in a Prompt 3.3 Saving and Sharing Prompts in Copilot Prompt Gallery 3.4 Scheduling Prompts

5 Chapter 4: Managing Conversations and Notebooks

4.1 Finding Previous Conversations 4.2 Renaming and Deleting Chats 4.3 Adding a Conversation to a Copilot Notebook

6 Chapter 5: Creating and Managing Microsoft 365 Copilot Agents

5.1 Agent Store vs. Creating a New Agent 5.2 Creating an Agent from a Template 5.3 Configuring an Agent That Has Knowledge 5.4 Configuring Instructions, Capabilities & Suggested Prompts 5.5 Sharing an Agent with Team Members

7 Chapter 6: Drafting and Analyzing Business Content

6.1 Creating a New Document from a Prompt 6.2 Generating a Document from an Existing Document 6.3 Generating a Management Summary 6.4 Moving Data & Insights Between Microsoft 365 Apps 6.5 Drafting Communications with Copilot in Outlook

8 Chapter 7: Meetings, Collaboration and Personalization

7.1 Using Copilot for Meetings: Setup & Live Intelligence 7.2 Meeting Recap, Transcripts & Catch-Up 7.3 Copilot Pages for Collaboration 7.4 How Copilot Uses Memory & Instructions

2.3 Recognizing & Mitigating Risks to Sensitive Data

Key Takeaways Explain This Topic AI

Microsoft 365 Copilot enforces strict Security Trimming: it executes strictly within the authenticated user's Entra ID context and cannot access, index, or surface content the user does not have explicit permissions to view.

Microsoft Purview sensitivity labels provide persistent classification and cryptographic encryption (Rights Management Services); Copilot honors these protections and applies label inheritance to derivative content.

Data Loss Prevention (DLP) policies prevent sensitive enterprise data (such as PII, credit card numbers, or proprietary IP) from being processed, summarized, or leaked via generative AI chats.

Enterprise oversharing—caused by overly broad SharePoint permissions like 'Everyone except external users'—is the leading risk factor when deploying Copilot, as AI makes previously obscure files instantly discoverable.

Remediation of oversharing requires a defense-in-depth approach: Restricted SharePoint Search, Purview Information Barriers, Entra Access Reviews, and Just-in-Time least-privilege permissions.

Last updated: August 2026

2.3 Recognizing & Mitigating Risks to Sensitive Data

When deploying generative AI across an enterprise, data security, regulatory compliance, and access control are paramount concerns. Business leaders frequently worry: Will Copilot expose confidential executive compensation data to all staff? Will our proprietary intellectual property be used to train public models? How do existing data classification policies apply to AI-generated text?

Microsoft 365 Copilot addresses these concerns through a comprehensive security architecture that integrates deeply with Microsoft Entra ID , Microsoft Graph , and Microsoft Purview . Understanding how access controls, sensitivity labels, and security trimming function is essential for the AB-730 certification.

Microsoft Purview Integration & Sensitivity Labels

Microsoft Purview Information Protection allows organizations to discover, classify, and protect sensitive data across cloud services, apps, and endpoints. Microsoft 365 Copilot natively respects and enforces Purview classifications. +───────────────────────────────────────────────────────────────────────────+

  • | Purview Sensitivity Taxonomy |
  • | |
  • | [1. Public] ── Non-sensitive data; free external distribution |
  • | [2. General] ── Standard business data; internal distribution |
  • | [3. Confidential] ── Sensitive business data; restricted access + RMS |
  • | [4. Highly ── Critical intellectual property / financial data; |
  • | Confidential] strict encryption, watermarking, no forwarding |
  • +───────────────────────────────────────────────────────────────────────────+
  • Sensitivity Label Inheritance

When Copilot generates new business artifacts (such as drafting a Word proposal, summarizing a PowerPoint deck, or answering a Teams chat), it enforces sensitivity label inheritance :

Single Source Inheritance : If Copilot creates a summary from a document labeled Confidential - Finance , the newly generated document automatically inherits the Confidential - Finance label and its associated encryption policies.

Multi-Source Inheritance (Most Restrictive Wins) : If Copilot references multiple documents to synthesize an executive report (e.g., one General document and one Highly Confidential document), Copilot automatically applies the most restrictive sensitivity label ( Highly Confidential ) to the resulting output.

Encryption Preservation : Rights Management Services (RMS) cryptographic encryption keys are preserved. Users who lack the cryptographic rights to decrypt a source file cannot have Copilot summarize it.

Data Loss Prevention (DLP) Policies

Microsoft Purview Data Loss Prevention (DLP) policies inspect content in real time to prevent the accidental or malicious exfiltration of sensitive information types (SITs)—such as Social Security Numbers, credit card numbers, medical records, or proprietary project codenames.

Prompt Inspection : DLP policies evaluate user prompts before they reach the LLM. If a prompt contains prohibited sensitive data, DLP can block the prompt or alert compliance administrators.

Output Filtering : DLP evaluates Copilot's generated responses. If an output contains sensitive data that violates organizational policy, the response is blocked or redacted.

Sensitivity Label Governance Matrix

Label Tier Typical Content Encryption / RMS Copilot Processing Behavior Label Inheritance Policy

Public Marketing brochures, public press releases None Fully accessible for grounding across public sites Output defaults to Public unless blended with higher tier

General Internal memos, team meeting notes, company news Optional Accessible to all authenticated tenant users Output labeled General

Confidential Vendor contracts, departmental budgets, client lists Yes (AES-256 RMS encryption) Accessible only to authorized group members; decrypted in-memory Output automatically inherits Confidential label + encryption

Highly Confidential M&A due diligence, executive compensation, trade secrets Mandatory (Strict user-level RMS + watermarking) Accessible strictly to named individuals; blocked if export rights restricted Output strictly inherits Highly Confidential; export/copy blocked

The Threat of Enterprise Oversharing & Remediation

While Copilot's security trimming ensures users only see files they have permission to access, it exposes a critical organizational vulnerability: Enterprise Oversharing . +───────────────────────────────────────────────────────────────────────────+

  • | The Oversharing Dilemma |
  • | |
  • | [Historical Reality: Security by Obscurity] |
  • | - File: "2026_Executive_Bonuses.xlsx" stored on intranet SharePoint. |
  • | - Permission: "Everyone except external users" (Open Read). |
  • | - Reality: Nobody knew the URL, so the file was effectively "hidden." |
  • | |
  • | [Modern Reality: Copilot Semantic Discovery] |
  • | - Employee Prompts: "What are the executive compensation targets?" |
  • | - Copilot: Indexes all accessible files, finds the open file instantly, |
  • | and outputs a complete summary in seconds! |
  • +───────────────────────────────────────────────────────────────────────────+
  • Why Oversharing Occurs

In many organizations, SharePoint sites and OneDrive folders have historically been configured with overly broad permissions, such as:

Sharing links set to "Anyone with the link" or "Everyone in the organization" .

Departmental document libraries configured with access granted to the "Everyone except external users" default group.

Public Microsoft Teams channels used to store confidential human resource or financial spreadsheets.

Before generative AI, organizations relied on security by obscurity —assuming that because an employee did not know a file existed or where it was stored, the data was safe. Copilot eliminates obscurity by semantically indexing all accessible enterprise data, making overshared files instantly discoverable via simple natural language queries.

Strategic Remediation Framework

To prevent oversharing incidents prior to and during Copilot deployment, organizations must implement a multi-layered remediation framework: ┌─────────────────────────────────┐

  • │ Oversharing Remediation Pillars │
  • └────────────────┬────────────────┘
  • │
  • ┌───────────────────────────┼───────────────────────────┐
  • ▼ ▼ ▼
  • ┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐
  • │ 1. Restricted │ │ 2. Entra Access │ │ 3. Purview Info │
  • │ SharePoint Search│ │ Reviews & JIT │ │ Barriers & DLP │
  • └─────────────────┘ └─────────────────┘ └─────────────────┘
  • │
  • ▼
  • ┌─────────────────┐
  • │ 4. Site Access │
  • │ Governance Audit│
  • └─────────────────┘

Restricted SharePoint Search : Administrators can configure Restricted SharePoint Search to limit organization-wide search and Copilot grounding to a curated list of up to 100 approved SharePoint sites, preventing Copilot from indexing uncurated long-tail sites while permissions are audited.

Microsoft Entra Access Reviews & Just-in-Time (JIT) Access : Regularly prompt site owners to audit and attest to group memberships, automatically revoking access for users who no longer require it under Least Privilege principles.

Purview Information Barriers : Establish communication and discovery firewalls between conflicting departments (e.g., preventing investment banking teams from accessing research analyst data, ensuring regulatory compliance).

SharePoint Site Access Governance : Audit and eliminate broad sharing groups ("Everyone except external users"), converting open sites to private group-backed sites with designated business owners.

Realistic Business Scenario: The Executive Compensation Incident

Scenario : Fabrikam Manufacturing deploys Microsoft 365 Copilot to 2,000 employees. On day two of the rollout, an entry-level factory supervisor enters the prompt: "What are the salary bands and bonus multipliers for senior management this year?"

The Exposure : Copilot returns a detailed table outlining executive compensation, bonus percentages, and equity grants.

The Root Cause Analysis :

Did Copilot experience a security breach or hallucination? No.

Did Copilot bypass Entra ID security permissions? No.

Investigation revealed that an executive assistant had uploaded Executive_Compensation_2026.xlsx to a SharePoint site originally created for general corporate benefits. The site permissions were set to "Everyone except external users (Read)."

The Governance Remediation :

Immediate Action : The compliance team applied a Purview Highly Confidential sensitivity label with restricted RMS encryption to the file, immediately cutting off access to unauthorized users.

Systemic Action : The IT administrator enabled Restricted SharePoint Search to contain Copilot's indexing boundary, while launching tenant-wide Entra Access Reviews to purge all legacy "Everyone" permissions from departmental sites.

Key Data Governance Concepts for AB-730 Candidates

  • Important
  • Critical Exam Takeaways on Data Protection

Security Trimming is Absolute : Copilot results are always filtered by the calling user's permissions.

Sensitivity Labels Stick : When generating derived content from labeled files, Copilot applies label inheritance (highest classification wins).

Encryption is Preserved : RMS-protected documents maintain encryption throughout AI interactions.

Fix the Permissions, Not the AI : Oversharing is an underlying SharePoint/access governance issue that must be remediated through proper access controls, not by blaming the AI tool.