To address AI risks, policymakers and technologists around the world have embraced the concept of AI red-teaming — the adversarial testing of AI models. But AI red-teaming is such a new concept that few people know how to conduct such tests. The discipline of AI red-teaming very much remains under development, but as AI systems are rolled out far and wide, this discipline is only going to become more important to make sure that AI tools work as intended. On this episode of Safe Mode, host Elias Groll sits down with Josh Harguess, the AI security chief at Cranium AI who is one of the few bona fide experts in the field, to dive deep on how to carry out AI red-teaming. CyberScoop reporter Derek Johnson also joins the show to discuss an Israeli information operation targeting audiences in the United States and Canada.