From Prompts to Plans: Security and Safety Testing for Agentic AI (Dec. 17th)
Thursday, December 17th, 2026: 2:00 PM to 3:00 PM
ServiceNow's Jason Stanley explains how to test agentic AI systems for security and safety risks using threat modeling and automated tools.
Agentic AI systems that plan, act and adapt introduce new errors and attack vectors that blur the line between safety failures and security breaches.
Safety and security testing by AI labs is often general-purpose and not tied to threat modeling for specific use cases, so it rarely reflects the risks of real deployments.
Jason Stanley, Head of AI Research Deployment at ServiceNow, describes an approach that starts with threat and risk modeling based on how agents are actually used.
He then covers an automated AI red teaming pipeline and DoomArena, an open-source security and safety testing framework for agentic AI.