CyberWire Daily · Software Engineering & Infra · November 2026
Forbes, of Booz Allen Hamilton, describes his team's lab tests of frontier AI models against industrial control networks. He says the pace was something they had not seen before.
I know you and your colleagues at Booz
Allen have been doing research on some of these frontier AI models and how they affect OT environments. Can we start off with an overview and maybe what prompted that research?
…across eight controlled scenarios. And in each case, the divine test objective of the model was achieved. So what does that mean? It means that we use these models on scenarios, had them navigate unfamiliar industrial systems, and they identified assets and vulnerabilities and were able to progress into control environments in eight different scenarios. Let me just give you one specific example. In one test, a model progressed from perimeter compromise to actions inside the control network in just over 16 minutes. So that was a game-changing revolution for us to see that the pace at which the AI agents can act on the OT network was really something that we had not seen before. Is it fair
to say that the results were surprising, as you say, the speed at which they were able to do the things that you
tasked them with doing? Well, you know, when we look at OT networks, we had to say to ourselves that in many ways, we're not necessarily identifying new vulnerabilities. I mean, OT networks typically have a life cycle of 20 to 30 years. So we expect going in that as opposed to IT, Dave, that gets…
Are you optimistic? Are you confident that we're heading towards a place where the defenders could have the upper hand here?