When AI acts security has to answer: Microsoft Security for AI | BRK336
Yina Arenas, Greg Schellenberg, Alym Rayani, and Dareen Radwan present a lifecycle approach to AI security, connecting agent governance with identity, data protection, runtime controls, and threat response.
Securing AI agents across their lifecycle
This Microsoft Ignite 2026 session examines the security challenges AI agents introduce beyond traditional SaaS governance. It focuses on three customer priorities: visibility into AI environments, control over agent access and actions, and the ability to respond to threats at machine speed.
Microsoft’s approach spans Azure, AWS, Google Cloud, and local environments. The session describes a layered security architecture using Microsoft Entra, Defender, Purview, Intune, and Sentinel to protect AI applications and agents, govern access and data, enforce network controls, and contain threats.
Discover, contain, protect, and adapt
The session organizes agent security into a four-part framework:
- Discover: Establish visibility into agents through the Agent 365 registry and dynamic agent risk scoring for cloud and local agents.
- Contain: Govern agent identities and permissions, including an Entra demonstration of dynamic permissions and Snowflake access controls.
- Protect: Apply Defender, conditional access, and runtime risk policies to AI agents.
- Adapt: Adjust security controls as agent risks change.
A KPMG case study covers managing hundreds of thousands of AI agents with Agent 365 and Perception, bringing the discussion from individual agent controls to governance at scale.
Session chapters
- 00:00: Securing AI agents and balancing innovation with control
- 01:00: Why AI agents create security challenges beyond traditional SaaS governance
- 04:00: Customer priorities: visibility, control, and machine-speed response
- 05:00: Cross-platform AI security across Azure, AWS, Google Cloud, and local environments
- 06:00: Layered architecture using Entra, Defender, Purview, Intune, and Sentinel
- 09:00: Discover, contain, protect, and adapt framework
- 10:30: Agent 365 registry and dynamic risk scoring for cloud and local agents
- 17:30: Entra agent identities and dynamic permissions demonstration with Snowflake access controls
- 26:00: Agent protection and adaptation with Defender, conditional access, and runtime risk policies
- 41:00: KPMG case study on Agent 365 and Perception
Resources
Session: BRK336 · Microsoft Ignite 2026 · Security · Intermediate breakout · English (US)