AI agents are spreading rapidly across organizations: from Copilot Studio to Azure AI Foundry, from the Microsoft 365 Agent SDK to third-party solutions.

An increasing number of organizations now have a growing portfolio of agents over which IT and security teams have limited visibility or control.

At the same time, these agents are typically not protected against emerging threats, while both the variety and volume of such threats are increasing significantly. In many ways, we are still living in the Wild West of AI agents, and it is time to start taming it using the tools already available to us.

In this session, we will review recent real-world agent attacks and security failures, examining what happened, which safeguards were missing, and what organizations could have done differently to reduce risk using our current agent security and governance toolkit.

During the session, we dive into Microsoft's technical solutions for AI agent governance, security, and oversight, along with the risks they help mitigate:
* The Agent 365 control plane: visibility, management, and protection of AI agents
* Entra ID: dedicated identities for agents in Entra ID, access management, network traffic filtering, and security controls
* Defender: real-time protection for agent sessions (threat protection)
* Purview: visibility and governance of sensitive data handling, auditing, logging, and forensic investigation

Ultimately, we will tackle the burning question: Is Agent 365 licensing a hard necessity for secure production agents?

This session is well-suited for security practitioners, IT decision makers and anyone responsible for the secure deployment and operation of AI agents in their organization.