Dawn Song: When AI Becomes the Hacker and the Defender

Dawn Song, Professor of Computer Science at UC Berkeley, joins Ann on this week’s episode of Afternoon Cyber Tea to explore how AI is rapidly reshaping the threat landscape. Dawn shares research from projects like Cyber Gym, demonstrating how frontier AI models are increasingly capable of identifying vulnerabilities, including zero-day exploits and generating proof-of-concept attacks, while also helping defenders automate patching and strengthen resilience.   The conversation examines adversarial AI, agentic systems, data privacy risks, and the need for science and evidence-based AI policy. Dawn closes with an optimistic vision for the future: using AI-driven formal verification to build provably secure systems and move beyond the traditional cat-and-mouse security model.  Resources:  View Dawn Song on LinkedIn    View Ann Johnson on LinkedIn   Related Microsoft Podcasts:   Microsoft Threat Intelligence Podcast   The BlueHat Podcast    Uncovering Hidden Risks           Discover and follow other Microsoft podcasts at microsoft.com/podcasts         Afternoon Cyber Tea with Ann Johnson is produced by Microsoft, Hangar Studios and distributed as part of N2K media network.