Overview
Learn about three emerging categories of Rogue AI and essential security measures in this 39-minute LASCON conference talk. Explore the security challenges that arose from rapid LLM adoption and the growing threat of excessive AI agency. Discover detailed insights into Accidental, Subverted, and Malicious Rogue AI categories, along with specific defense strategies for each type. Gain practical knowledge from the OWASP Top 10 for LLM Applications, focusing on preventing Excessive Agency. Master the implementation of a defense in depth (OSI) model to counter potential evasions and understand how to develop an advanced Zero Trust maturity model for robust AI security. Walk away equipped with the skills to identify intentional AI behaviors and implement effective risk detection and mitigation strategies when artificial intelligence systems deviate from intended behaviors.
Syllabus
Josiah Hagen - Three Categories of Rogue AI and How to Mitigate your Risk
Taught by
LASCON