AUSA News: Army Sprinting to Field AI Agents to Counter Autonomous Cyberattacks

WASHINGTON, D.C. — The Army has initiated three projects to develop artificial intelligence capabilities to enhance the service’s cyber defenses, with the first of these efforts expected to award a pair of contracts later this week, officials said Oct. 12.

The Army Rapid Development of Cyber Defense Systems program came following a tabletop exercise with 14 C-suite executives from both the defense industry and multinational corporations already using AI for cyber missions, Brandon Pugh, principal cyber advisor to the secretary of the Army, said during a panel at the Association of the United States Army’s Annual Meeting and Exposition.

The exercise generated 20 possibilities the Army could pursue, with the service selecting three lines of effort to keep the program’s scope manageable, Pugh said during a media roundtable following the panel.

The service released the solicitation for the first effort, known as Project Griffin, in late August. “The premise was, if we have thousands of autonomous attacks being launched against us, what is the art of possible to respond?” Pugh said.

The Army already has an AI-driven monitoring platform called Panoptic Junction detecting malicious activity at a roughly 90 percent success rate, he said. The goal of Project Griffin is to take action against those threats so a “human operator doesn't have to get on keyboard instantly, perhaps it can do a series of response actions that we give it guardrails to do in response to that adversary behavior.”

The Project Griffin solicitation was only open for a week, and the service received 108 submissions, Pugh said. The Army then narrowed that down to 10 companies and is planning on awarding contracts to two of them this Friday.

Meanwhile, the service is still working through what exactly those “response actions” the AI agents will take against adversary threats will look like, said Lt. Gen. Christopher Eubank, commander of Army Cyber Command.

Service leaders are actively engaging with industry “to understand what they're doing in the take action space and what it really means, what are the guardrails involved in that, so that we can take what they've learned and apply it to Project Griffin as we roll it out,” Eubank said during the roundtable.

The aim is to start with the Army’s portion of the Defense Department Information Network and then expand it as the service learns how to use the capability, he said.

“Honestly, I think for us, it's going to be, are we as humans ready to let the agent actually take action?” Eubank said. “You think that for how many years a human has been the person making the patch, remediating something, and so we're walking ourselves through what does it mean to have an agent do that work, and then how do you check that agent's work? So, a lot still to be done in this space.”

Pugh noted the Army kept the solicitation broad upon the recommendation of industry partners.

“We wanted to do one or two narrow actions, and [the industry members were] like, ‘We highly recommend you don't do that, because we can guarantee you we're doing things in this space that you're not even thinking were possible, and you might not have the appetite or the risk tolerance to do that. But at least don't box us in to only doing two or three capabilities. Let us give you the art of the possible,’” he said.

The process for developing response actions will be iterative, he added. “Once these contracts are awarded, perhaps we start small, and we only have it do a couple take action elements. We test that in our environment, see how it operates in our network, and that could evolve over time to doing even additional actions.”

Along with Project Griffin, the Army is pursuing two additional efforts focused on agentic deception and “purple teaming” capabilities, Pugh said.

The goal of the agentic deception agents would be to lure adversaries away from the service’s “crown jewels” and redirect them to a portion of the network with less important data so the malicious actors spend “down their tokens, if they are leveraging AI against us, away from parts of the network” the Army actually cares , he said.

Meanwhile, purple teaming capabilities would allow the service to test its cyber defenses in an automated fashion, an activity that has traditionally been a very manual process, he said.

While Pugh did not provide specifics on when the solicitations for the two projects would be released, he said he is pushing the Army to go faster.

“My challenge to the acquisition community has actually been like, ‘OK, you're saying it can be open here and awarded like a month or two later. … How do we now cut that in half and have it open in like a week and then close?’” Pugh said. “That's actually what we did with Griffin. We greatly drove down the actual timeline which would be traditional for an acquisition process, without jeopardizing any step of the acquisition process. It just meant a lot of people have put a ton of time and attention into this and quickly vetting 108 submissions.”

Pugh added he is continuously advocating for the Army to make additional investments in AI for cyber applications, noting senior leaders have shown strong interest in the technology.

“I'd love to see a scenario where we continue to sprint to bring these capabilities in that industry largely has, so that way we don't have to just start from scratch,” he said. “We can take an existing commercial solution and perhaps fine-tune it and bring it here in a period of weeks, not years.”

Topics: Army News , Robotics and Autonomous Systems