Artificial intelligence is transforming nearly every aspect of modern life, but it still has limitations. Thanks to Hollywood and video games, there’s a misconception that AI-agents like Tony Stark’s “Jarvis” or Master Chief’s “Cortana” are just moments away from implementation. But bringing AI into the military battlespace comes with challenges that civilian applications don’t face.
Military AI must protect sensitive information because that data cannot risk exposure through cloud systems or network queries. And in hostile territory, even a radio ping can give away a unit’s position, potentially with lethal consequences.
Download the SAN app today to stay up-to-date with Unbiased. Straight Facts™.
Point phone camera here
That raises a critical question: How can warfighters benefit from AI without compromising their security or revealing their location?
A new ‘Manhattan Project’ moment
Tyler Saltsman, CEO and co-founder of EdgeRunner AI, argues the stakes could not be higher.
“AI’s are the next frontier. These things can be more powerful than nuclear weapons,” Saltsman said. “So, we have to –– all of us –– sort of just come together like it’s the Manhattan Project again. We have the big tech folks, we have the small startups like us. We all have different core competencies. I think we all need to work together to safeguard our nation.”
EdgeRunner AI is part of a small but growing group of defense-focused technology companies working to address the security, connectivity and operational challenges of using AI in combat environments.
Central to EdgeRunner’s mission is the idea of an air-gapped AI assistant. That means using AI models that operate locally, on any device, without any internet or satellite connection. This keeps data secure and avoids broadcasting a soldier’s position.
Saltsman called it a “better version of ChatGPT that never needs the internet and that’s military specific.” Meaning, there is an AI agent programmed specifically for each military occupation specialty, or MOS.
Instead of a single generalized model, EdgeRunner builds adapters for each military role. For example, a logistics officer receives very different support than a medic, a pilot or a tank operator.
These personalized AI agents are taught using military service doctrine, unit-standard operating procedures, and technical manuals. All tasks are processed locally on the user’s device.
Examples of AI on the edge
Saltsman, who served as an Army logistician, described a moment during an Eastern Europe deployment when he wished such a tool existed. His team needed to repackage explosive ordnance after a truck malfunctioned.
“We had a truck go down, and now we had to repackage explosive ordnance from one truck to another truck,” he said. “Well, the question is, ‘Will this exacerbate the explosives and take out my convoy? Or is it completely fine? Can these mix together?’ Right away, an AI would be able to know, ‘Yeah, that’s good. You’re good to go’ or ‘You’re not.’”
Reducing cognitive load in high-stress moments
Beyond technical questions, Saltsman said EdgeRunner’s systems help operators think clearly in chaotic conditions.
“Imagine you’re running a mission and you’re getting shot at, or things are hitting the fan, you know? Sometimes it’s hard to think in those high-stakes environments,” Saltsman said. “This can help you by reducing the cognitive load on the warfighter so they can move forward.”
According to the company, this value proposition has helped it gain traction across the Defense Department, including with the Army, Air Force, Space Force, and Special Operations Command. Saltsman said EdgeRunner systems are already being used overseas with SOCOM units.
This fall, EdgeRunner AI conducted field evaluations with the 10th Special Forces Group at Fort Carson and with the 82nd Chemical, Biological, Radiological and Nuclear (CBRN) Company units. Saltsman said the tests validated the technology’s ability to quickly parse documents, generate operational plans, draft orders, and provide suggestions based on uploaded unit doctrine.
“Now you can just load it right into your PC. And the AIs will speak to those documents,” he said.
The goal is to give entire units a common operating picture. “If that group all have the same AIs, and it’s not like a Frankenstein solution of different vendors, you get that baseline truth. And now you can all march in the same direction.”
Partnerships across the DoD and Big Tech
EdgeRunner AI’s model relies heavily on Meta’s Llama architecture, which it fine-tunes into military-specific agents.
“Meta is great because we don’t have to pre-train these models from scratch,” Saltsman said. “They provide the engine. We build everything else.”
EdgeRunner AI also signed a Cooperative Research and Development Agreement with the Air Force Research Laboratory to support NIPRGPT — a generative AI system running on the Air Force’s nonclassified network.
EdgeRunner was also designated as an awardable vendor in the Department of Defense Tradewinds Solutions Marketplace and is contributing to efforts to bring on-device AI aboard U.S. Navy ships.
While AI can speed decision-making and improve accuracy, Saltsman stressed repeatedly that it should not replace human judgment.
He put it bluntly, “AIs make smart people smarter, and they make dumber people dumber. Now, the key is knowing this isn’t the silver bullet. You need to verify.”
That includes understanding biases in training data. Saltsman warned that many commercial models are basically trained by studying the internet, including social media comment sections, with little filtering.
“These biases, they’re not always pro-America or pro-Western civilization. We have a lot of just crazy ideologies out there that are in these research labs, and they’re building AIs in their own image, and you can’t do that,” Saltsman said.
EdgeRunner claims it solves this by making its training corpus visible to customers, a level of transparency that major labs do not currently offer.
A race the US cannot lose
Saltsman argues that to further develop AI, America must move quickly but carefully. He believes China is already ahead in some areas of AI development, including training efficiency and hardware utilization.
“We need to make sure that we keep our foot on the gas. We don’t want to fall behind in the AI race, just like we fell behind with the DJI protocols with drones.”
And the stakes extend beyond pure technology. Saltsman warned that if China were to seize Taiwan, the global semiconductor supply chain — and by extension, the future of AI — could be destabilized.
