Skip to content

WhyDSPM.md

Latest commit

 

History

History
189 lines (126 loc) · 11.4 KB

WhyDSPM.md

File metadata and controls

189 lines (126 loc) · 11.4 KB

Why DSPM for AI? — Value Proposition Guide

This guide explains what each automation step does and why it matters for your organization. Use it to understand the business value of the accelerator and communicate it to stakeholders, security teams, and compliance officers.

Understanding the Two Governance Paths

Critical Distinction: This accelerator supports two different AI governance scenarios with different underlying mechanisms. Choose the path that matches your workload.

Scenario What's Governed How It Works Tags to Run
M365 Copilot & 3rd-Party AI Apps Microsoft 365 Copilot, ChatGPT, other browser-based AI Purview DSPM captures prompts/responses and routes them to user mailboxes for retention, eDiscovery, and DLP m365
Microsoft Foundry & Custom Agents Azure OpenAI, Cognitive Services, custom agentic solutions Defender for AI monitors via diagnostic telemetry to Log Analytics; Purview registers data sources for classification defender, foundry

Key Differences

Aspect M365 Copilot Path Microsoft Foundry Path
Where prompts are stored User mailboxes (Exchange Online) Log Analytics workspace (diagnostic logs)
Retention mechanism M365 retention policies Log Analytics retention settings
eDiscovery Yes (via Purview eDiscovery) No (requires custom export)
DLP enforcement Yes (Purview DLP policies) No (Content Safety blocklists instead)
Threat detection Purview Insider Risk + Defender Defender for AI only
Licensing required M365 E5 or E5 Compliance Azure subscription

Choose Your Path

  • Govern M365 Copilot and browser-based AI apps → Run m365 tag from a desktop with MFA
  • Protect Microsoft Foundry projects only → Run defender,foundry tags (no M365 license needed)
  • Cover both paths → Run all tags, understanding that governance mechanisms differ per workload

The Bottom Line

What this accelerator does for you:

If you're protecting... The accelerator...
M365 Copilot / 3rd-party AI Wires your AI interactions into Purview so prompts live inside the same compliance boundary as email and Teams—subject to retention, eDiscovery, and DLP
Microsoft Foundry Enables Defender for AI threat detection, streams diagnostics to Log Analytics, and applies Content Safety guardrails—but prompts do NOT flow to mailboxes
Both Configures both paths, giving you unified visibility in Defender for Cloud while maintaining the distinct governance mechanisms

Every automation step corresponds to official Microsoft guidance—nothing proprietary or black-box. The scripts exist to remove toil while respecting your organization's approval gates.

Official Microsoft Learn References

These links validate the automation steps below:

Step-by-Step Value Breakdown

Step 1: Prepare the Configuration Spec

What Happens Why It Matters
Copy spec.dspm.template.json to spec.local.json and populate with your tenant, subscription, and resource IDs. Optionally regenerate the template when the schema changes. The spec is your single source of truth. Every script consumes it, ensuring consistent governance across Purview, Defender, and Azure AI workloads. One file drives everything.

M365 Copilot Path (Steps 2-4)

These steps apply to M365 Copilot and 3rd-party AI apps. Skip to Step 5 if you're only protecting Microsoft Foundry.

Step 2: Enable Unified Audit and M365 Prerequisites

What Happens Why It Matters
Run ./run.ps1 -Tags m365 to confirm Unified Audit is enabled, activate DSPM for AI switches, and log which prerequisites passed. Audit is the M365 compliance safety net. Without it, Copilot interactions won't be captured. This step guarantees the M365 foundation is solid.

Step 3: Create the Secure Interactions Policy

What Happens Why It Matters
The automation creates the "Secure interactions from enterprise apps" policy—the same one-click policy Microsoft Learn calls DSPM for AI - Capture interactions for enterprise AI apps. This is M365-specific. Microsoft routes M365 Copilot prompts/responses to user mailboxes so they inherit your existing retention, eDiscovery, and Communication Compliance controls. Microsoft Foundry prompts do NOT flow through this path.

Step 4: Complete Portal-Only M365 Policies

What Happens Why It Matters
The script prints a checklist of policies that must be finished in the Purview portal (Communication Compliance, Insider Risk). Some M365 features are still UI-only. Calling them out prevents a false sense of "automation complete" and keeps compliance owners in the loop.

Microsoft Foundry Path (Steps 5-8)

These steps apply to Microsoft Foundry and custom AI solutions. They use Defender for AI and diagnostics, NOT mailbox-based capture.

Step 5: Register Foundry Resources with Purview

What Happens Why It Matters
The automation enumerates your Microsoft Foundry workspaces and projects, registers them as Purview data sources, and enables the Secure data in Azure AI apps and agents toggle. This connects Foundry to governance—but differently than M365. Purview can classify data in Foundry resources, but prompts are NOT routed to mailboxes. Telemetry flows to Defender and Log Analytics instead.

Step 6: Verify Configuration

What Happens Why It Matters
A verification script checks Foundry registration, Defender plan status, and diagnostic settings, then summarizes the state. Trust but verify. You see exactly what succeeded and what's pending. This is the validation checkpoint before declaring success.

Step 7: Enable Defender for Cloud + Defender for AI

What Happens Why It Matters
The automation registers providers and enables the Defender for AI plan, reporting which AI services are now protected. Defender is where Foundry threat protection lives. It detects prompt injection, jailbreaks, and data exfiltration—but through diagnostic telemetry, not mailbox monitoring.

Step 8: Enable User Prompt Evidence & Purview Integration

What Happens Why It Matters
Scripts attempt preview APIs; if unavailable, they log the portal path to enable user prompt evidence and the Purview integration toggle in Defender for Cloud. This connects Defender to Purview for Foundry workloads. Without it, you can detect a malicious prompt but can't correlate it with Purview's data classification. Note: This is different from the M365 mailbox-based capture.

Both Paths (Step 9)

Step 9: Review Logs and Evidence

What Happens Why It Matters
The orchestrator prints success/failure lines and saves full transcripts under logs/governance/. Proof of work. You leave the session with a snapshot you can attach to a change ticket, email to auditors, or use as a diagnostic starting point if something fails later.

Summary: What You Get Per Path

Capability M365 Copilot Path Microsoft Foundry Path
Prompt/response capture Yes - To user mailboxes No - Diagnostic logs only
Retention policies Yes - M365 retention Partial - Log Analytics retention (configure separately)
eDiscovery Yes - Purview eDiscovery No - Requires custom export from Log Analytics
DLP enforcement Yes - Purview DLP policies No - Use Content Safety blocklists instead
Sensitivity labels Yes - Applied to captured content Partial - Applied to Foundry data sources (classification only)
Threat detection Yes - Insider Risk + Defender Yes - Defender for AI
Prompt injection detection Partial - Via Communication Compliance Yes - Defender for AI native capability
Required licensing M365 E5 or E5 Compliance Azure subscription (PAYG)

Who Should Care?

Stakeholder M365 Copilot Path Microsoft Foundry Path
CISO / Security Leadership AI interactions governed by same controls as email—no shadow AI visibility gaps Threat detection and posture management via Defender for Cloud
Compliance Officers Prompts subject to retention, eDiscovery, and DLP—audit-ready from day one Diagnostic logs available for export; no built-in eDiscovery
SOC / Security Operations Insider Risk and Communication Compliance alerts Defender for AI alerts in the same console they already use
AI/ML Platform Teams N/A (M365 managed by IT) Governance automated—focus on building, not configuring compliance
IT Operations One spec file, one run—deterministic M365 policy deployment One spec file, one run—Defender + diagnostics configured

Common Objections Addressed

Objection Response
"Unclear which AI apps are in use" M365 path: DSPM for AI discovers AI interactions across M365 Copilot and third-party AI apps accessed via browser. Foundry path: Define which Foundry projects to monitor in the spec.
"Microsoft Foundry only" Run defender,foundry tags only. No M365 license required. You get threat detection and diagnostics without the M365 overhead.
"Governance will slow down AI initiatives" The automation runs in minutes and doesn't block AI workloads—it observes and protects
"Existing DLP is enough" M365 path: Traditional DLP doesn't see AI prompts. DSPM extends existing policies to cover AI-specific data flows. Foundry path: Use Content Safety blocklists instead of DLP.
"This seems complex" The accelerator reduces 50+ portal clicks to a single azd up command with a config file
"Do Foundry prompts go to user mailboxes?" No. Only M365 Copilot/3rd-party AI interactions route to mailboxes. Foundry telemetry goes to Log Analytics via diagnostics.

Next Steps

  1. Review the README for deployment options
  2. Follow the Deployment Guide for step-by-step setup
  3. Check the Troubleshooting Guide if you hit issues
  4. Review Cost Guidance to understand billing implications