Frontier AI is changing what is possible in OT cybersecurity attacks. Not necessarily by creating entirely new attack techniques, but by dramatically reducing the time, specialized knowledge, and human effort required to execute them. This session moves beyond speculation to share findings from controlled testing of frontier AI models in a realistic, multi-vendor OT environment.

Drawing on a series of controlled testing scenarios conducted in a realistic OT lab environment, this session will explore what happened when leading frontier models were introduced into an unfamiliar industrial environment and given attack targets. Could they understand the architecture, navigate trust boundaries, adapt when their initial approaches failed, and translate digital access into actions affecting physical equipment? The results challenge several long-held assumptions about the protection industrial complexity, proprietary technologies, and specialized OT knowledge provide.

Attendees will gain a grounded view of demonstrated frontier AI capabilities, where current defenses retained leverage, and how operators can prepare by constraining attack paths, hardening critical assets and trusted applications, detecting consequential changes in industrial behavior, and strengthening containment and recovery.