The rapid advancement of AI Agents is creating a new wave of automation and intelligent system operations. Unlike traditional chatbots that are limited to conversational interactions, AI Agents are capable of reasoning, using tools, executing commands, and directly interacting with systems such as email, web browsers, terminals, and enterprise workflows. While these capabilities can significantly improve operational efficiency, they also introduce a wide range of new cybersecurity risks.

This presentation will focus on analyzing some of the most prominent cybersecurity threats emerging within today’s AI Agent ecosystem from a practical enterprise deployment perspective. Topics will include scenarios such as supply chain prompt injection through malicious dependencies or skills, indirect prompt injection via email and enterprise workflows, as well as risks arising when AI Agents are granted access to internal systems and operational tools.

In addition to threat modeling and live demonstrations, the session will also introduce several approaches to help organizations deploy AI Agents more securely, including AI Gateway, Prompt Firewall, sandbox runtime environments, skill/MCP ecosystem governance, and methods for detecting Shadow AI Agents through EDR and system telemetry.

Through practical case studies and real-world demonstrations, the presentation aims to help organizations gain greater visibility into the emerging attack surface of AI Agents and explore how AI can be adopted in a more controlled and secure manner within enterprise environments.