AI Is Already on Your Network. Is Your Access Model Ready?
This webinar from AppGate explores the security risks posed by AI and agentic AI systems in enterprise networks and how zero trust principles can address these challenges. The session covers three main risk categories for AI models: data exposure and model training concerns, model poisoning where malicious actors influence training data, and model backdooring with trigger-based attacks. The speakers discuss how agentic AI systems, which combine language models with tools and external system access, create new security threats including tool misuse, identity abuse, and rogue agent behavior. They present real-world examples like an AI agent that accidentally deleted three months of production data at a car rental company. The solution framework involves three layers: Observe (gaining visibility through an AI gateway), Control (applying proper permissions and policies), and Protect (using small language models for runtime protection). The webinar includes interactive polling about current AI usage and emphasizes the need for proper identity management, least privilege access, and continuous monitoring in AI deployments.
