This interview podcast focuses on the evolving landscape of AI security in 2025. The host begins by discussing predictions from the Association for Computer Machinery's FACT conference and cybersecurity expert Chuck Brooks, highlighting concerns around AI's speed versus safety and security, and the need for robust testing and ethical considerations. The main interview features Marco Figueroa, program manager for the ODIN Gen AI bug bounty program, who details his work identifying vulnerabilities in AI systems, including a specific jailbreak allowing access to a ChatGPT code interpreter's file system. Figueroa emphasizes the importance of creative thinking in AI security research and the ODIN program's focus on rewarding researchers for finding and reporting these vulnerabilities. Listeners gain insights into current AI security challenges and the opportunities for creative problem-solving in this rapidly evolving field.
