Dr. Mikel Rodriguez is the MITRE AI Fellow and a member of the National Artificial Intelligence Advisory Committee. A veteran of both the Pentagon and Silicon Valley, he previously served as a Senior Technical Leader at Google DeepMind, where he led AI Red Teaming and offensive cyber capability evaluations on Gemini. In the public sector, he was a special government employee at the Defense Innovation Unit and was the Founding Technical Architect for the DoD’s Project Maven and the Director of AI Research at MITRE. He holds a Ph.D. from the University of Central Florida and is a recognized expert on AI assurance and national security strategy.
Presentation Title:
Where Frontier AI Security and Cyber Are Heading: A Field Report Through a National Security Lens
Presentation Abstract:
In the last year, AI systems have scaled their offensive cyber capabilities faster than the field expected. Tasks that were out of reach a few months ago now run at lower cost and higher reliability: discovering zero-day vulnerabilities, writing working exploits, executing multi-stage cyber campaigns end to end. These capabilities are no longer confined to closed labs. Open-weight models, including recent Chinese releases, trail frontier ones by months, not years.
This is a working field report on where AI security and cyber stand today: what the capability is doing for attackers and for defenders, what current instruments can and cannot measure, and where this is heading as autonomous agents and multi-agent systems arrive in real deployments. It draws on time inside a frontier lab and inside the national security missions now built around these systems.
For national security, the strategic variable is pace. The US is fielding AI aggressively in the high-stakes domains where national security lives. Real mission gains are already showing up. So are new vulnerabilities, at exactly the moment offensive capability is scaling on the other side. Closing the defensive gap faster than an adversary can open the offensive one is the actual competition.
Genesis lands inside that competition. Chartered as the American Science and Security Platform, it holds several of the hardest open problems in AI security inside the workflows this community will build over the next few years. Whether those problems get answered here, or first by a peer nation working the same capability curve, will be decided inside those workflows, not around them.