Understanding Data Loss Prevention M365
Data loss prevention in M365 protects email, files, chat, and AI tools. See where Microsoft Purview stops and how Strac adds discovery and redaction.
Data loss prevention in M365 is a set of policies and controls that stops sensitive data from reaching the wrong person. It covers PII, PHI, PCI data, credentials, and confidential files. Microsoft delivers its native version through Purview.
A policy looks for a data type and applies an action. An email with a card number can trigger a warning, a block, or an alert. Strac's guide to Microsoft 365 DLP policies covers the setup.
Old oversharing used to stay hidden. A forgotten folder was low risk because nobody went looking for it.
Copilot changes that. It respects each user's permissions, which is exactly the problem: one prompt can surface a salary spreadsheet the user never knew they could open. Permissions tell you who can reach a file. They do not tell you what the file contains.
Outside AI agents add a second path. Through the Microsoft 365 MCP server, tools such as Claude, Cursor, and ChatGPT can read mail, files, and chats for a user. The Model Context Protocol (MCP) returns that content to the agent, so sensitive data can leave the tenant in a single tool call.

Purview covers Exchange, SharePoint, OneDrive, Teams, and endpoint devices. It can warn, block, encrypt, and notify. In 2026, Microsoft also added controls for Copilot prompts, protections for Copilot Cowork, and a preview for data at rest in connected apps such as Box and Google Workspace.
The remaining gaps are practical:
A policy tells you the rule was broken; remediation removes the exposure.
The leak is usually an ordinary shortcut, and each surface has its own version.

Strac inspects the message body and every attachment, including text inside images through OCR. It then redacts the sensitive value in place and leaves the rest of the email readable. The original stays in an encrypted vault for authorized roles.
Redaction fixes the risky value; a block stops the whole conversation.
Strac works alongside Purview and independently of your Microsoft license tier. The Office 365 DLP integration is API-based, so email and file coverage does not require software on every laptop.
Strac starts with DSPM. It runs data discovery and classification across existing M365 content, labeled or not, and shows who can reach it. You can add custom data elements for company-specific identifiers.
Labels show what someone already classified; discovery finds what nobody did.
From there, Strac applies four remediation actions:
Strac stops the sensitive value before it reaches the prompt, so there is nothing to clean up later.
For agents, MCP DLP inspects each tool response, redacts sensitive values, applies approval rules, and logs the call. Each agent acts with the requesting user's identity, so Microsoft 365 permissions still decide what comes back.
The same policies extend to Google Drive, Gmail, Box, and other integrations. Audit logs support reviews for HIPAA, PCI DSS, SOC 2, and ISO 27001.
Strac closes the gap between a written policy and the data itself. Start with the Office 365 DLP guide, or read AI DLP for tools beyond Copilot.

👉 Related reading:
M365 DLP now has to cover people, Copilot, and outside agents. Identity, prompt, and permission controls all fail eventually; the data layer is the backstop. Purview sets the rules for Microsoft, and Strac finds and fixes the sensitive data across M365 and beyond.
Book a demo to see what is exposed in your tenant.

Not exactly. Purview is Microsoft's product for creating DLP policies. M365 DLP is the wider job of finding, controlling, and fixing sensitive data across Microsoft 365, which can also involve tools outside Purview.
No. Many teams keep Purview as the native baseline for Microsoft policies. Strac adds discovery of unlabeled data, in-place redaction, and coverage for other SaaS apps, browsers, endpoints, and agents. See Microsoft Purview DLP for the comparison.
Yes. Redaction and vaulting keep the work moving because they remove only the sensitive value. The email still sends, the file stays useful, and authorized people can retrieve the original when they need it.
No. Detection can miss unusual data, and permissions are rarely perfect. That is why the data layer matters: when sensitive values are redacted and excess access is removed, a mistake or compromised account has far less to expose.
MCP DLP inspects the content passing between AI agents and Microsoft 365. It can redact sensitive data, require approval for risky actions, and log each tool call. Start with the MCP DLP pillar for the full picture.
.avif)
.avif)
.avif)
.avif)
.avif)

.gif)

