What's changed: Added per-section figures (cert-figure-retrofit). New AB-900 Chapter 2 (Domain 2 "Data protection and governance": Purview = Information Protection/sensitivity labels, DLP, Insider Risk, Communication Compliance, DSPM for AI, Data Lifecycle Management/classification; Copilot data security = within existing permissions/Microsoft Graph grounding/overlap of permissions-Purview-Defender/responsible AI; risk identification and oversharing = Compliance Manager/Data-Activity Explorer/DLP alerts/eDiscovery Content search/data access governance report/SharePoint Advanced Management restricted site access)
2.2Data security of Copilot
Understand how Microsoft 365 Copilot accesses data, how Microsoft Graph influences its responses, how Copilot is protected against risks via permissions and controls in Microsoft 365/Purview/Defender, and responsible AI principles.
The key to Copilot’s security is the principle that Copilot only operates within the data a user can already access. Copilot does not create new access; it inherits existing permissions. That is exactly why oversharing or wrong permissions surface through Copilot.
2.2.1Microsoft Graph and responses
Copilot uses Microsoft Graph to access organizational context—email, files, chats, calendar—and, respecting the user’s permissions, grounds its responses in relevant information (grounding). Graph is why Copilot can answer "in the context of your organization’s data." Because access scope determines response scope, permission design and classification affect both response quality and safety.
2.2.2Protection via permissions/Purview/Defender, and responsible AI
Copilot risks are mitigated by overlapping controls: (1) Microsoft 365 permissions (Copilot cannot use data the user cannot access), (2) Microsoft Purview (sensitivity labels, DLP, DSPM for AI to classify/protect/visualize sensitive data), and (3) Microsoft Defender (threat detection/XDR). It also follows Microsoft’s responsible AI principles (fairness, reliability & safety, privacy & security, inclusiveness, transparency, accountability), addressing risks like hallucination and harmful output.
Watch the mix-ups: (1) Copilot does not create new access—it inherits existing permissions, so oversharing surfaces via Copilot. (2) Microsoft Graph grounds responses but never exceeds the user’s permissions. (3) Protection is not a single feature but the overlap of permissions + Purview + Defender.
Common: Copilot data behavior → correct statement. E.g., "can Copilot see all company data for anyone?" = no (within the user’s permissions); "what if there is oversharing?" = it can surface via Copilot; "why can Copilot answer in your org’s context?" = grounding via Microsoft Graph; "how to visualize sensitive-data risk?" = Purview DSPM for AI.
2.2.3Section summary
- Copilot operates only within the user’s existing access; oversharing surfaces via Copilot
- Microsoft Graph grounds responses in org context (never exceeding permissions)
- Protection = M365 permissions + Purview (labels/DLP/DSPM for AI) + Defender; follows responsible AI
Sign in to track progress — Log in.
Quick check
(just a quick review)Q1. Which correctly describes how Microsoft 365 Copilot accesses data?
Q2. Through what does Copilot access context to ground responses in your organization’s data?
Q3. What is the effect on Copilot if an organization has oversharing (over-broad sharing)?
Q4. Which is NOT part of the overlapping controls that mitigate Copilot risk?
Q5. Which framework provides principles like fairness, transparency, and accountability for using AI including Copilot?
Q6. Which Purview capability best visualizes and manages the risk of sensitive data Copilot handles?