AI agents are often described as tools for office work, software tasks, or online errands. Scout AI is applying the same broad idea to a far more consequential setting: military systems that can search for and destroy targets in the physical world.
From digital assistants to combat systems
Scout AI is part of a new wave of defense technology startups trying to adapt advances from large AI labs for military use. Instead of using agents to write code, answer emails, or buy products online, the company is training them to operate with self-driving vehicles and lethal drones.
In a recent demonstration at an undisclosed military base in central California, Scout AI placed its technology in charge of a self-driving off-road vehicle and a pair of lethal drones. The mission was to find a hidden truck and destroy it with an explosive charge.
Colby Adcock, Scout AI’s CEO, described the company’s goal in stark terms. “We need to bring next-generation AI to the military,” he said. “We take a hyperscaler foundation model and we train it to go from being a generalized chatbot or agentic assistant to being a warfighter.”
How the demonstration worked
The demonstration began with a command entered into a Scout AI system called Fury Orchestrator:
Fury Orchestrator, send 1 ground vehicle to checkpoint ALPHA. Execute a 2 drone kinetic strike mission. Destroy the blue truck 500m East of the airfield and send confirmation.
A relatively large AI model with over a 100 billion parameters interpreted that instruction. According to the source article, Scout AI uses an undisclosed open source model with its restrictions removed. The system can run on a secure cloud platform or on an air-gapped computer on-site.
That larger model then acted as an agent. It issued commands to smaller, 10-billion-parameter models running on the ground vehicle and drones. Those smaller models also operated as agents, sending instructions to lower-level AI systems that handled movement.
After receiving the order, the ground vehicle drove along a dirt road between brush and trees. A few minutes later, it stopped and launched the two drones. The drones entered the area where the target was expected to be, identified the truck, and one drone’s AI agent ordered it to fly toward the target and detonate an explosive charge just before impact.
Why autonomy is the central claim
Scout AI argues that the value of its system lies in greater autonomy. Adcock says this is what separates the company’s approach from older autonomous systems. “This is what differentiates us from legacy autonomy,” he said.
He said legacy systems “can't replan at the edge based on information it sees and commander intent, it just executes actions blindly.” In practical terms, Scout AI is presenting its agents as systems that can interpret orders, process what they encounter, and adjust actions based on the mission.
That same feature is also what makes the technology sensitive. A system that has room to interpret instructions may be more flexible, but it also raises questions about unintended outcomes. The source article notes that the idea of an AI system freely interpreting orders creates concern, especially when the system is connected to weapons.
The military opportunity and the risks
Many policymakers believe AI will be central to future military dominance. The combat potential of AI is also one reason the US government has sought to limit the sale of advanced AI chips and chipmaking equipment to China, although the Trump administration recently chose to loosen those controls.
Michael Horowitz, a professor at the University of Pennsylvania and former Pentagon deputy assistant secretary of defense for force development and emerging capabilities, said defense startups should be testing AI integration. “That's exactly what they should be doing if the US is going to lead in military adoption of AI,” he said.
But Horowitz also pointed to the gap between a compelling demonstration and a system that can be trusted in military conditions. Large language models are inherently unpredictable, and AI agents can misbehave even when assigned relatively benign tasks like ordering goods online.
Cybersecurity is another major barrier. Horowitz said proving that such systems are robust from a cybersecurity standpoint may be especially hard, and that would be required for widespread military use.
What must be proven next
The US and other militaries already have systems capable of autonomously exercising lethal force within limited parameters. Critics warn that off-the-shelf AI could make autonomy easier to deploy more widely and with fewer safeguards.
Arms control experts and AI ethicists also warn that AI-controlled weapons create new complexities and ethical risks. One example in the source is the question of whether AI would be required to decide who is and is not a combatant.
The war in Ukraine has already shown how cheap, off-the-shelf hardware like consumer drones can be adapted for deadly combat. Some systems already feature advanced autonomy, although humans often make key decisions to ensure reliability.
Collin Otis, cofounder and CTO of Scout AI, says the company’s technology is designed to follow the US military’s rules of engagement and international norms like the Geneva Convention. Adcock says Scout AI already has four contracts with the Department of Defense and is competing for another contract to build a system for controlling a swarm of unmanned aerial vehicles. He also said it would take a year or more before the technology is ready for deployment.
For now, the central question is not whether Scout AI can stage a dramatic test. It is whether AI agents can become reliable, secure, and controlled enough for military use. Horowitz put the caution plainly: “We shouldn't confuse their demonstrations with fielded capabilities that have military-grade reliability and cybersecurity.”