Safeguard data across your apps and agents development cycle | OD829

Arpitha Dhanapathi explains how to build data security and compliance into AI app and agent development from day one, using Microsoft Purview as a unified policy and governance layer. The session covers common data leakage paths, practical guardrails, and how teams can move from prototype to production without reinventing controls.

Overview

Why AI apps and agents increase data risk

• Developers move quickly, but generative AI scenarios can unintentionally overshare sensitive data.
• Enterprise concerns include:
  • Data leakage through prompts, responses, logs, and tool/plugin calls
  • Compliance and policy violations when prototypes reach production without governance

Mapping enterprise concerns to developer challenges

• The session connects common enterprise risk questions (what data is exposed, where it flows, who can access it) to day-to-day developer realities:
  • Rapid iteration and experimentation
  • Multiple AI stacks (Microsoft and third-party)
  • Difficulty aligning with existing enterprise controls without slowing delivery

Microsoft Purview as a foundational trust layer

• Microsoft Purview is presented as a way to provide:
  • Integrated governance
  • A simplified developer experience for applying enterprise policies
  • A unified approach to protecting data across apps and AI agents

Paths to trusted AI (Microsoft and third-party scenarios)

• The session describes integrating Purview protections across:
  • Microsoft AI scenarios
  • Third-party AI scenarios
• Focus is on using a consistent policy layer rather than building one-off controls per app.

Demo: AI observability and Data Security Posture Management in Purview

• Demonstrates AI observability concepts and the Purview portal experience for Data Security Posture Management (DSPM).

Related Purview capabilities highlighted

• Insider Risk Management
• Compliance Manager
• Communication Compliance

Real-time Data Loss Prevention for Microsoft Foundry

• Introduces real-time DLP for Microsoft Foundry, positioned as a guardrail to reduce leakage risk during AI development and deployment.

Extending protection to local agents

• Discusses extending Purview-based protection to local agents, aiming to keep the same trust and policy approach across environments.

Session metadata

• Event: Microsoft Build 2026
• Session code: OD829
• Level: Intermediate
• Topic area: Cloud platform & data
• Resource link: https://aka.ms/build26-next-steps