Army Officials Plan to Reduce Cyber Risks of Artificial Intelligence
The U.S. Army last month initiated a 100-day plan to explore ways to reduce risks associated with artificial intelligence (AI) algorithms and intends to follow that up with a 500-day plan beginning this summer.
Army officials held a live and virtual media roundtable with reporters April 22 to discuss the service’s AI plans. The discussion covered AI-related small business innovation research (SBIR) awards and Project Linchpin, which will serve as a pathway for AI algorithms developed for programs of record. Army officials participating in the discussion included Young Bang, principal deputy within the office of the Assistant Secretary of the Army for Acquisition, Logistics and Technology (ASA ALT); Matt Willis, director of Army prize competitions and the Army Small Business Innovation Research (SBIR) program; and Bharat Patel, who leads Project Linchpin for the Program Executive Office-Intelligence, Electronic Warfare and Sensors.
“We’re going to set the conditions for an AI implementation plan for ASA ALT specifically in our programs,” Bang said. The plan includes three “sprints” to “codify setting the conditions” for ASA ALT and the Army. It also includes five “levels of effort” indicating the Army’s focus areas, along with “timelines associated with the sprints” that outline goals and milestones.
The Army intends to rely heavily on industry for AI algorithms, Bang noted. “The Army recognizes that we’re not going to be doing algorithms and model development and training better than industry. We want industry to do that. And they do that extremely well, so we want to adopt a lot of that.”
But risks include the potential for poisoned data sets and adversarial attacks, such as Trojans. Protecting against those risks is easier in a controlled, trusted environment owned by the Army or the Defense Department, Bang added. “And so yes, we’re going to do all that. We’re going to put in all the security components and bake that in from a development standpoint, just like you would do software. But this really looks at how can we adopt third-party or commercial vendors’ algorithms, right into our network, into our programs so we don’t actually have to compete with them.”
The Army recognizes that we’re not going to be doing algorithms and model development and training better than industry.
So, the Army is implementing a “layered AI risk reduction or risk framework,” to identify those risks. Bang promised the service will find a better name for the framework and will work with industry to categorize risks and explore tools and processes to reduce those risks so that the service can adopt AI more rapidly.
Willis told reporters that in fiscal year 2025, the Army will could see a 40% increase in SBIR programs aimed at AI innovations, potentially up to $150 million, and much of that will be aimed specifically at Project Linchpin. “This really demonstrates our commitment to innovation and AI and how small businesses across the country can certainly contribute to the Army,” Willis said. “We currently have about $115 million in active SBIR projects that are aligned to AI and for the most part Project Linchpin.”
In an interview days prior to the media roundtable, Lareina Adams, program manager for terrestrial sensors, said the algorithms developed under the third-generation forward looking infrared (3GEN FLIR) program might benefit Project Linchpin. “We’ve made quite a few advancements in terms of algorithms that we can feed into Project Linchpin to help that effort mature. So, I think going forward, it will need to be closely linked with what they’re doing,” Adams said.
Christina Bates, strategic communicator for the terrestrial sensors program manager, added that the 3GEN FLIR’s high-quality images could help teach Project Linchpin’s AI algorithms. “A big part of the value out of the third-gen—beyond the immediate operational need—is that it produces the images of the requisite caliber, against which, for example, Linchpin can start to develop more robust models to engage in machine learning and ultimately, algorithm generation as they look to building out that algorithm pipeline,” Bates said. “It’s kind of the garbage in, garbage out idea, right? The better the inputs are in terms of imagery, the better the algorithm learns, and the more robust and accurate that algorithm will be.”
During the media roundtable Patel confirmed the collaboration between 3GEN FLIR and Project Linchpin. While the Tactical Intelligence Targeting Node is the primary use case for Linchpin, the 3GEN FLIR is another, he reported. “Third-gen FLIR, think tank-on-tank kind of action. They’re trying to identify things in that sensor [that] we want to be able to apply. We want to be able to apply some AI to help them identify things that the human eye can’t see.”
However, since 3GEN FLIR is a new sensor, it has not yet generated enough images for Linchpin to work with, Patel added. “That sensor has not seen enough objects in it for us to actually do some AI techniques, so one of our goals would be to get that thing out there in different exercises and get data.”
Some upcoming SBIR opportunities will be related to 3GEN FLIR and Linchpin, Patel suggested. “That gets after some of the SBIRs that we’re going to be focusing on in the next six to eight months. [There] will be the model training possibilities for that particular sensor through the Project Linchpin pipeline into a program.”
The Army owns the 3GEN FLIR imagery, so presumably, it would be less risky than some other data.
Bang indicated the Army has been a AI leader within the U.S. military and its efforts have drawn interest from the other services and the Office of the Secretary of Defense for potential partnerships or sharing lessons learned.
He also said the Army will provide reporters with a two-page description of the AI risk reduction plan, so this story could be updated when that happens.
