Why AI Changes the Risk Model for Application Security

James Wickett discusses how the rapid integration of AI, particularly LLMs, is reshaping application security by introducing probabilistic behavior, new data access patterns, and operational risks that legacy tools fail to capture. He outlines the limitations of traditional shift-left approaches—especially noisy, non-actionable findings—and emphasizes the need for clear AI risk definitions, reference architectures, and practical controls for threats such as prompt injection and excessive agency so teams can maintain speed without sacrificing trust or security.