Federal AI Security Depends on Vendors and Tight Budgets

The U.S. Army Corps of Engineers is directing capital toward modern infrastructure and data systems as it builds artificial intelligence capabilities. Cybersecurity experts say agencies must keep people involved in AI decisions. Federal agencies are testing and deploying tools across government networks and critical infrastructure, but budgets and computing capacity limit what they can use.

Public Systems, Capital Investment

The Corps’ AI strategy has three pillars: accelerating adoption, educating and training the workforce, and building enduring, tested, trusted and verified capabilities. Matt Parks, the chief artificial intelligence officer for the Corps of Engineers, said cybersecurity must run through all three. He described targeted capital investments in modern infrastructure, resilient systems, stronger data governance and Army Vantage, which the Corps uses to operationalize data and make it accessible.

Parks said the work also requires a culture of cybersecurity awareness. The Corps is integrating cyber into its AI ecosystem of capabilities, authoritative data and modern IT infrastructure, he said. Workforce education sits alongside investment in infrastructure and technology.

The Corps uses digital twins in an operational technology monitoring initiative that covers critical infrastructure, including hydropower dams and locks. Parks said the agency faces constant cyber attacks against those facilities and is trying to mitigate the risk. The Corps is considering AI-powered threat detection to sort thousands of cyber risk alerts, many of them incorrect, so staff can focus on the important ones.

At the Department of Agriculture, the cybersecurity and privacy operations center uses AI tools to monitor agency networks, conduct continuous monitoring and threat analysis, and provide centralized awareness, analytics and reporting. Rudy Rojas, an information technology manager and security specialist at the department, said he reviews reports and decides whether to whitelist or blacklist activity. The department isn't giving AI total control. A human remains involved in cybersecurity decisions.

People Remain in the Loop

Rojas said the system is working, and vendor-based tools are moving quickly toward AI-powered predictive analytics. AI also supports the USDA’s zero trust effort. 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 assess “securing from AI, securing with AI and securing AI,” according to Scott Bowman, the technical lead for cyber-physical systems at INL. The COGNITO program examines consequences before AI adoption, including effects on workflows and systems that will depend on the technology, and how critical AI may become to a system.

Bowman raised questions about reliance on an application programming interface, a cloud API or Codex. What happens if a service becomes unavailable, or a budget or token limit is reached? INL also examines AI tools’ code and architectures through what Bowman called a “mechanistic analysis.” He said this work can help people understand how AI can be used to understand AI, while warning that people will never understand every dependency well enough to feel comfortable.

Procurement Under Constraints

Chris Cardran, a federal security advisor for defense and intelligence at Splunk, said he often recommends a smart machine model approach because AI models are complex. The process involves creating a model, training it on a customer’s data and searches, extracting information, grading it and putting it into operation. These models handle specific tasks and need less computing power than a large frontier model. Teams can take them offline for updating or retraining before grading them and returning them to operation.

Cardran said smart machine models can support operational technology security and machine-based questions about cyber logs and metrics. Kevin Dorsey, the director of federal solutions development at Optiv+Clearshark, said customers are seeking bespoke large language models that meet governance requirements and perform needed work. Agencies need models that can run in data centers with limited space, he said, while budgets for purchases such as GPUs remain constrained. Dorsey cited AI assistance in detecting and collating information on customer tactics, techniques and procedures or standard operating procedures as a practical use.