DARPA’s Andrew Carney on AIxCC’s quest for truly autonomous AI

Safe Mode Podcast

Greg Otto talks with Andrew Carney, project manager at DARPA, about the AIxCC competition. With the finals set to be held in August during DEF CON, they discuss how these autonomous systems fared in detecting and remediating vulnerabilities, as well as the key lessons learned from live exercises and the semifinals. The conversation highlights DARPA's vision to merge formal software engineering with large language models to dramatically reduce software vulnerabilities and explains the scientific and engineering advances still needed to achieve this goal at scale. We also examine the challenges of safeguarding critical infrastructure, particularly when so much depends on open-source projects maintained by volunteers, and consider the impact of AI on patch deployment, code verification, and sustainable defense. In the reporter chat, Greg talks with Matt Kapko about a story debunking the 16 billion password hack.