Army Cyber Command is scaling several artificial intelligence pilots to make cybersecurity insights more actionable this year, according to its commander.
Speaking at TechNet Augusta 2025 on Tuesday, Lt. Gen. Maria Barrett said the data ARCYBER collects off the Army network will serve as “ammunition” for about a dozen AI projects.
These projects involve AI assistants, agentic AI and human-machine teaming, and they vary in maturity across the enterprise. ARCYBER has shifted its focus to faster analysis, as its security capabilities have advanced.
“Now we have the ability to see adversary reconnaissance before the attack,” Barrett said. “That drives an entirely different set of decisions for me.”
Click the banner below to start implementing smarter security.
Introducing GHOSTCREW and Expanding Panoptic Junction
The Army has prioritized sending ARCYBER and 11th Cyber Battalion red teams into real-world events to test weaknesses in how they operate, Barrett said.
Now they’re getting a new tool, GHOSTCREW, which supplies red team operatives with AI simulations to inform their decision-making. GHOSTCREW predicts attack vectors and recommends responses, leaving operatives to make the final call — an example of the human-machine teaming the Army is increasingly embracing.
ARCYBER is also expanding Panoptic Junction, its AI-driven continuous monitoring platform, into a 12-month production pilot.
“It’s part of a White House [defensive cyber operation] for AI,” Barrett said.
Panoptic Junction analyzes system architectures to learn how they were built, and examines threat intelligence to prioritize vulnerabilities for monitoring. The prototype boasted an 87% success rate at detecting malicious activity.
Unlock Exclusive Cybersecurity Insights
Complete the form below to be redirected to CDW's exclusive proprietary research report on Cybersecurity. Once the form is submitted, you’ll be opted into our Security email stream.
The Army’s Streamlined DevSecOps Process Saves Time to Deployment
Over the long term, the Army wants to use AI to build code libraries and for autonomous Digital Control Organizations and electronic warfare actions, Barrett said.
The Army’s honed DevSecOps process already scans every code commit (the process of saving source code changes in a shared repository) because of the threats posed by the ubiquitous use of open-source libraries. Rather than slowing down software deployments, the Army Cyber Technology and Innovation Center (ArCTIC) performed its last security update to CamoGPT — the branch’s machine learning-enabled equipment management platform — in hours instead of months, Barrett said.
To better understand the software in its inventory, the Army is also automating its software bill of materials, a comprehensive inventory of components.
Click the banner below to keep up with FedTech as an Insider.
Accidental Leaks and Avoiding AI’s ‘Whitaker Effect’
AI is not without its pitfalls. Soldiers should be wary of accidental leaks when using applications (some department-approved) to do work faster because they may have AI running in the background that trains itself with uploaded data per the end-user license agreement, says Hansang Bae, public sector CTO at Zscaler.
A second problem is what Bae calls the “Whitaker effect.” Named after an infamously inbred American family, it occurs when AI trains itself with the increasingly available AI-generated content found online. The result is the perpetuation of common AI errors, such as the generation of images of six-fingered hands.
“We’re poisoning AI little by little,” Bae says. “It’s getting dumber as it’s getting smarter.”
Bae’s solution? Government needs to create a tagging taxonomy that differentiates organic, human-generated data from AI-generated data. The Department of Defense could then mandate that the taxonomy be followed and that no AI-generated data be used in training.
Until the problem is addressed, the Army must rely on AI with caution, particularly when soldiers’ lives are on the line.
“There's clearly a place for AI,” Bae says. “But fluid combat situations are probably not the place to start.”
To learn more about TechNet Augusta, visit our conference page. You can also follow us on the social platform X at @FedTechMagazine to see behind-the-scenes moments.
