Video summary

Secure the age of AI: Redefining trust, data and access

Main summary

Key takeaways

Technology

Summary

Core theme

Organizations must “secure data and access in the AI era.” Traditional security models—mainly designed around applications/files and human users—are insufficient for AI workloads that include prompts, agents, connectors, and data flows that existing controls can’t reliably see or govern.


New security drivers (scale + risk)

  • Exponential growth of identities and endpoints
    • Includes both human users and non-human entities/AI agents.
  • Explosion in enterprise data
    • Claim: 90% of the world’s data was created in the last 2 years.
  • Attack surface expands
    • Every identity/endpoint becomes a potential entry point and a “cache of enterprise data.”
  • AI-related incidents are common
    • 70% of organizations report AI incidents.
  • Non-human identities will dominate access requests
    • By 2029, 60% of secure access requests are expected to come from non-human identities (AI agents), requiring access enforcement that goes beyond people.

Why access control must change

  • Access can’t be granted/denied only at the application layer
    • AI requires controls spanning prompts and granular workflows.
  • Access can’t be limited to human identities
    • Agents must be governed similarly to employees.
  • Data protection must be real-time while data is moving
    • Not only “at rest.”

Proposed model: Unified, adaptive “access fabric” (Zero Trust)

The approach moves beyond siloed identity/network security into an “access fabric” that secures every digital interaction.

Three Zero Trust–driven principles

  1. Contextual access
  2. Continuous evaluation
  3. Connected real-time protections

How it works conceptually

  • Identity determines who/what should have access.
  • Network access continuously validates trust/session security based on identity signals.
  • The goal is zero-trust, continuous evaluation across interactions involving humans, machines, and agents, with adaptive governance and enforcement.

Microsoft Entra + Microsoft Purview integration

Entra (identity/access)

  • Provides the identity/access side, including:
    • Identity
    • Conditional access
    • Insider risk context

Purview (data security)

  • Provides data security controls that protect data in real time as it travels, not just while stored.

Together: connect access decisions to data protection

  • Access decisions are informed by data sensitivity.
  • Enforcement adapts based on context at the speed and scale AI operates.

“One home” concept

  • Entra manages access policies tied to identity/network/session trust.
  • Purview manages how data is secured across the digital estate and where it travels (endpoints, apps, third parties).

Data movement scenarios driving tighter protections

  • Data accessed via enterprise apps (e.g., Microsoft 365) and third-party apps (Workday, Salesforce, line-of-business apps).
  • Data can be:
    • Downloaded
    • Copied to removable media
    • Printed
  • In the AI era, data may be shared to:
    • Unmanaged browsers
    • Consumer-grade AI apps
    • This can send “crown jewel” data to risky third parties.

Therefore: the emphasis is on network-level data protections integrated with Entra.


New / announced capabilities for AI and data-in-motion

Network data security integration (public preview)

  • Secures data in motion when shared to:
    • Shadow AI
    • Unmanaged SaaS apps
    • Personal cloud storage over the network
  • Goal: prevent sensitive data leakage

Future direction (planned deeper demos)

  • Entra + Purview walkthroughs showing the integration end-to-end.
  • Expanded protections for AI agents, including:
    • Agents running locally on devices
    • Agents running in the cloud

Packaging / availability

  • These capabilities are stated as available in Microsoft 365 E7:
    • Advanced data security via Purview
    • Identity/access via Entra

Main speakers / sources

  • Diana Visar — Product Marketing Manager, Microsoft Entra (host)
  • Chenade OdonovanMicrosoft Entra Vice President of Product
  • Matili DandichMicrosoft Purview General Manager of Product

Original video