AI, Cybersecurity, and Securing Model Weights with Miles Brundage and Chris Rohlf

In this episode of The AI Policy Podcast, Gregory Allen hosts Miles Brundage and Chris Rohlf to discuss the intersection of cybersecurity and AI. Chris shares his experience in vulnerability research and exploit development, highlighting the shift from manual processes to automation and the impact of AI. Miles discusses his work on malicious uses of AI and the increasing importance of cybersecurity in the AI landscape. The conversation explores the current state of cybersecurity, the offense-defense balance, and the potential of AI to benefit defenders. They delve into the challenges of securing AI models, the value of stolen model weights, and the need for robust cybersecurity practices in leading AI labs. Finally, they discuss policy recommendations for the government and companies to promote AI adoption and strengthen cybersecurity defenses.