At last August’s DEF CON computer security conference, more than 2,000 people tried their hands at breaking some of the world’s most advanced AI models. That event was the largest-ever public red-teaming event of large language models, and since then policymakers are continuing to look to red-teaming as a key tool in responsibly deploying AI systems. The data scientist Rumman Chowdhury was one of the organizers of the Generative AI Red Teaming Challenge at DEF CON, and on this episode of Safe Mode she sits down with host Elias Groll to discuss the lessons of that event. CyberScoop reporter AJ Vicens also joins the show to discuss a potentially severe supply chain attack involving the business analytics firm Sisense.
