Pentagon Ready to Let AI Make Some Decisions

“NATIONAL DEFENSE MAGAZINE” By Laura Heckmann

“This marks the Pentagon’s first foray into integrating AI agents in and across military workflows to provide advanced decision-making support systems for Defense Department leaders”

____________________________________________________________________________

“In the wake of the release of the Chinese artificial intelligence tool DeepSeek — branded a “wake-up call” for the United States by President Donald Trump — the Defense Department is working with industry to harness the next breed of AI in the global race.

In particular, the Pentagon is investing in agentic AI as increasingly sophisticated tools emerge not only across the United States but globally. Agentic AI operates with more independence than generative AI applications that rely on prompts.

San Francisco-based Scale AI announced in March the Defense Innovation Unit had awarded the company a prototype contract for its Thunderforge initiative — a project designed to utilize artificial intelligence in operational and theater-level planning, according to a DIU release.

Scale AI — along with its technology partners Anduril and Microsoft — will develop and deploy AI-powered solutions and custom agentic workflows that will be initially delivered to Indo-Pacific Command and European Command, a Scale AI release stated.

This marks the Pentagon’s first foray into integrating AI agents in and across military workflows to provide advanced decision-making support systems for Defense Department leaders, according to the Scale AI release.

Jason Droege, Scale AI’s chief strategy officer, said DeepSeek was a reminder that there is a global community, “accentuated in China, that is competing with us aggressively.”

The artificial intelligence race is also a race to write history, Droege said during a panel at the Air and Space Forces Association’s Warfare Symposium in March. As they are used today, most AI tools are a collection of the questions asked of them, reflecting clear biases of the country that produced them.

“If our collective memory is going into these models, and everything that comes out of the models has a lens of whatever government initiatives or bias or desire there is to put forward a history that is not favorable to the U.S., I think it’s absolutely alarming, and changing the hearts and minds … is a very, very big part of winning the war,” he said.

Removing human bias is one of the benefits of agentic AI, Droege said.

He described the technology as applications “that can pull in new data, live data, historical data and help you make decisions, take actions, surface decisions and actions to humans — if you want to do it safely — within our … military.”

For example, if China makes a gray zone move in the South China Sea, agentic AI could pull “any piece of new data that could be interesting or relevant … like, how do you take that move and then map it against every other known time that that happened?” he said. “How do you allow an operator to ask the next question? ‘Well, if they did this, what else did they do?’”

This type of analysis typically takes days or weeks — agentic AI can do it in seconds or minutes, he said. “And if you imagine that example … applied over thousands and thousands of people, many, many times a day, for every single input, every single sensor … around the world, that’s what an agentic AI system can do. It can process it all without human bias” — which is the goal of the Thunderforge initiative.

Droege said while he could not talk about specific use cases within Thunderforge, speed and access to data will be paramount, because “if data powers the effectiveness of our models, and models power the effectiveness of our actions and our military, then … there’s not a choice. And so speed is of the essence here.”

Air Force Maj. Gen. Steve Butow, military deputy of the Defense Innovation Unit, said cultural adoption of AI will be another critical piece of the initiative, and is usually “the hardest thing to change.”

“You can’t show up at the 11th hour at the fight and turn on AI and use it,” he said. “You have to train with it. You have to use it every day.”

Droege said advances in artificial intelligence have created a “very pivotal moment, not just for private enterprise, not just for consumers, but for the nature of how governments interact with each other, through policy, through military, through action. And it’s really important that we get this right.” 

Pentagon Ready To Let AI Make Some Decisions