Skip to content

BRK336 Breakout Tue 17 Nov 14:00 PST

When AI acts, security has to answer: Microsoft Security for AI

Yina Arenas, Dareen Radwan, Alym Rayani, Greg Schellenberg

Summary

From the recording / Microsoft AI summary

Agents were framed as an 'agentic insider' needing continuous control, using a know, contain, protect, adapt model across identity, data, endpoint, network and runtime. An Agent 365 registry (powered by Entra) feeds a dynamic agent risk score based on runtime risk, exposure and blast radius. Demos showed Entra dynamic permissions and conditional access blocking a risky agent. Stated as coming soon: a unified Perception experience, security-for-AI playbooks, Entra dynamic conditional access, and Defender posture for Bedrock, Gemini/Vertex and Anthropic agents; attack surface graphs in early December. KPMG reported about 230,000 agents, growing by about 1,000 a week (customer figure). Caveat: the speaker corrected 'private preview' to 'coming soon'; captions garble several names; no pricing or licensing was stated.

Session description

Microsoft's catalogue description

Learn how to accelerate AI innovation with confidence by securing AI across its lifecycle. Discover how to protect apps and agents, govern AI access and data, enforce network controls, gain visibility into AI risk, and contain threats through a unified security approach.

Demo timestamps

Microsoft's player does not support links to a specific time, so open the recording and skip to the time shown.

  1. Skip to 17:30

    Entra agent identities and dynamic permissions with Snowflake access controls

    Demo of Entra agent identities and dynamic permissions with Snowflake access controls, moving between explanation and a portal walkthrough.

Watch on-demand (opens in a new tab)

Resources named in the session

← All sessions