I served as a judge for the Critical Infrastructure Security Showdown (CISS), an international cybersecurity competition focused on operational technology (OT). Teams tackle security challenges leveraging physical critical infrastructure testbeds and digital replicas, putting their skills to the test in environments that reflect industrial systems.
This years environments included maritime, 5G network, and a power transmission. As you may expect, agentic AI was heavily used this year. Every team but one relied on LLMs to perform network reconnaissance and script generation. However, the teams steering and taking an active role with the AI performed the best by far.
It was not an easy CTF. 4 hours to perform reconnaissance, lateral movement, exploitation, and demonstrate some level of operational process control is not much time. It was amazing to see how the teams approached problems differently, and how some very experienced CTF players navigated the environment.
One of the biggest takeaways was the difference between how frontier models were leveraged in this competition, what they did well, what they did poorly, and how different it was from the claims made by the frontier labs leadership. No doubt AI will be leveraged in most cyber operations from this point forward. Fully agentic cyber operations impacting industrial systems, speaking ICS protocols, and manipulating industrial processes?
A claim not yet proven by the evidenc$$e.