The U.S. Army Corps of Engineers’ strategy for artificial intelligence revolves around three main pillars.
The USACE wants to accelerate AI adoption, educate and train the workforce, and build enduring, tested, trusted and verified AI capabilities.
Matt Parks, the chief artificial intelligence officer for the Corps of Engineers, said those pillars can’t be successful unless cybersecurity is woven throughout each of them.
“I’ve made it a real point to include cyber in everything that I do to make sure that the AI ecosystem triad of capabilities, authoritative data, and modern IT infrastructure are operating at a high, high level,” Parks said on the discussion How AI Is Reshaping Cybersecurity Strategy and Security Operations. “We’re making smart, targeted capital investments in modern infrastructures. Our systems are resilient by design rather than just some afterthought. Our data governance foundation is strong. We’ve made a large investment with Army Vantage to get all of our data operationalized so we can access that data, but that really comes with building a culture of cybersecurity awareness.”
An example of how the Corps of Engineers is integrating cyber, AI and mission requirements is in a monitoring initiative for operational technology (OT).
Parks said USAE is using digital twins to protect critical infrastructure that runs hydropower dams, locks and other similar facilities.
“We do get constant cyber attacks from bad foreign actors on those things. So we’re always trying to mitigate that risk,” Parks said. “We’re really thinking about, and it’s directly related to cyber, AI-powered threat detection. We, like probably everyone else, get thousands and thousands of cyber risk alerts that aren’t right. So how do you filter all that out? We’re trying to use AI to filter that out to get to the real ones and get less distracted from the ones that don’t mean anything so we can get focused on the ones that mean a lot.”
Three-part lens for AI and cyber
The use of AI to understand data to move toward a more proactive cyber mitigation environment is a common use case among public and private sector organizations.
Rudy Rojas, an information technology manager and security specialist at the Department of Agriculture, said their cybersecurity and privacy operations center is using AI tools to monitor agency networks, to do continuous monitoring and threat analysis to provide centralized awareness, analytics and reporting.
“These reports will go back to somebody like me. I’ll review those reports and determine whether we need to do a whitelist or whether we need to create a blacklist,” Rojas said. “But it’s something that we’re not allowing AI to solely have total control over. It’s something that we need to keep a human in the loop for the cybersecurity. So far, it’s working good. Our vendor-based tools are really moving headlong into utilizing AI to increase our predictive analytics.”
AI is also a part of USDA’s zero trust journey. Rojas said the agency’s chief information officer has made endpoint protections a priority to protect data and systems and manage software updates.
At the Idaho National Lab in the Energy Department, cybersecurity leaders are viewing the opportunities through a three-part lens: secure from AI, secure with AI and securing AI.
Scott Bowman, the technical lead for cyber-physical systems at INL, said one use case they are focusing on is using AI before implementing new cyber protections.
“We have a program called COGNITO that is before you adopt AI, understanding the consequences of it, not only of what it does to your workflows, but the systems that will become dependent upon it. How critical does AI become a function within that system?” Bowman said. “One thing that I witness myself, being a user of AI, is what happens whenever I don’t have that application programming interface (API) anymore or that cloud API? What happens when I don’t have Codex? What happens when you know the budget or token limit gets reached? For this specific use case, you have to use AI to get people to understand their processes really clearly and in greater detail.”
At the same time, Bowman said INL is securing AI tools by taking a deep dive into the code. He called it a “mechanistic analysis” of these models, where INL gets into the architectures of them.
“That is something where you can elevate people’s understanding of how you can use AI to understand AI because it’s so difficult to understand. There’s a lot of use cases wrapped into it,” he said. “For securing AI, the one thing that I really would like to bring home is that you won’t be able to understand all the dependencies to the point where you’re comfortable. You’re never going to get comfortable, and if you do get comfortable, then it’s probably not the right direction.”
Start small with SMMs
Chris Cardran, a federal security advisor for defense and intelligence at Splunk, said due to the complexity of AI models, he often recommends that his federal clients apply AI tools in a smart machine model (SMM) approach first.
“I can actually create my own model, train it on my own data using my own searches to be able to pull out and extract the information that I need to train it on, then grade it, and once it’s graded, I can then operationalize it,” Cardan said. “SMMs are for a very specific task. It doesn’t require as much horsepower, for lack of a better term, to run because it’s only doing one specific thing. It’s not this large frontier model that’s been trained on everything under the sun. I can keep that self-contained, and if I need to go back and update it or retrain it or add more context into it, I can. I just take it offline, retrain it, regrade it again, reoperationalize it, and I’m off and running. But that all keeps it in one location and self-contained.”
Cardran said the SMM approach also makes a lot of sense for protecting OT systems.
He said SMMs can support machine-based questions to understand cyber logs and metrics.
“You’re then training those models, like our time series based model that’s all about forecasting and automation, on, machine-based data and not cognitive data,” he said. “I think OT is probably one of the greatest places where you’re going to see some great benefit of AI.”
Kevin Dorsey, the director of federal solutions development at Optiv+Clearshark, said finding those practical use cases where bespoke LLMs can have the greatest benefits on mission success is major challenge and focus for their customers.
“What we’ve had to do is try to be creative with finding bespoke models that can do the job that we need and pass any AI governance that our customers require. We don’t have the luxury of, maybe call it luxury, going to Deep Seek and pulling in a Chinese model with our customers. So it’s finding models that can do the job, are small enough to run in a data center where we may have limited space and can still produce those same capabilities and provide our customers the results they’re looking for. That’s a big challenge,” Dorsey said. “As anyone who is trying to buy a GPU lately can tell you, budgets are constrained in this area, and so we’re trying to find the most practical use cases that we can support, whether it’s having our AI assist in the detection and collation of all the information, whether it’s tactics, techniques and procedures or standard operating procedures of a customer.”
Copyright © 2026 Federal News Network. All rights reserved. This website is not intended for users located within the European Economic Area.