At a US military base in central California, four-seater all-terrain vehicles navigate hillside trails in a training exercise dedicated to preparing AI models for conflict zones.
Scout AI, a startup founded in 2024 by Coby Adcock and Collin Otis, operates these autonomous military ATVs. Positioned as a “frontier lab for defense,” Scout AI announced a $100 million Series A funding round on Wednesday, led by Align Ventures and Draper Associates, following its $15 million seed round in January 2025.
TechCrunch was invited for an exclusive tour of Scout’s training operations at a confidential military base.
Scout is developing an AI model named “Fury” to command military assets, initially for logistical support, but expanding to autonomous weapons. CTO Collin Otis equates this development, which builds on existing LLMs, to the process of training soldiers.
“They start when they’re 18 years old, and sometimes they even start after college, so you want to start with that base level of intelligence,” Otis explained to TechCrunch. “It’s useful to start with someone who’s already made an investment and then say, hey, what do I have to do to teach this thing to be an incredible military AGI, versus just being a broadly intelligent AGI?”
Scout has secured $11 million in military technology contracts from organizations like DARPA, the Army Applications Laboratory, and other Department of Defense entities. Their technology is being used by the US Army’s 1st Cavalry Division at Ft. Hood in Texas, with an expectation for deployment in 2027.
Scout’s vehicles undergo rigorous testing in the base’s hilly terrain under the leadership of former soldiers. These operations are part of simulated missions.
As autonomous cars gain visibility in structured urban environments globally, managing unmarked trails off-road poses a new challenge. Otis, once at autonomous trucking company Kodiak, initiated Scout when he saw current systems couldn’t handle unpredictable war zones.
An autonomous ground vehicle controlled by Scout AI’s Fury model.
A new approach to autonomy
Scout explores Vision Language Action models (VLAs), based on LLMs, to control robotic systems. First launched by Google DeepMind in 2023, VLAs have enhanced startups like Physical Intelligence and Figure.AI, with Adcock sitting on Figure’s board. His experience led him to advocate bringing broader intelligence to the military’s autonomous vehicle fleet.
“If I handed you the controller of a drone right now and I strapped a headset on you, you could learn to fly that thing in minutes,” Otis remarked. “You’re actually just learning how to connect your prior knowledge to these couple little joysticks. It’s not a big leap. That’s the way to think about VLAs and why they’re such an unlock.”
Driving a Scout ATV around challenging trails, even without extensive ATV experience, proves the potential for developing general intelligence models. Scout trains its models using ATVs for a few weeks after civilian ATV trials.
Test drives reveal noticeable differences: autonomous control accelerates faster, hugs the correct side on wider trails, handles confusion with deliberate slowing, etc.
Though VLAs haven’t yet reached operational settings, their experimentation with soldiers is encouraged by technologists like Stuart Young, a former DARPA program manager. Scout’s full autonomy stack is complemented by deterministic systems and other AI types.
Young recently joined Field after DARPA’s RACER program, which challenged companies to develop high-speed, autonomous off-road vehicles, helping shape this sector akin to DARPA’s Grand Challenge with self-driving cars. Along with companies like Field AI and Overland AI, Scout participates in this rapidly evolving space.
Initial applications of ground autonomy, as per Scout’s vision, focus on automated resupply, like transporting water or ammunition to distant posts, or convoys where a crewed truck leads several autonomous vehicles. Brian Mathwich, an active-duty infantry officer serving as a military fellow at Scout, reflected on a night exercise in Alaska where autonomous support would have been beneficial.
Scout anticipates becoming a key software provider, enhancing military machines with an intelligence layer rather than constructing autonomous vehicles itself.
The expected flagship product, “Ox,” includes command and control software bundled with robust computer hardware (GPUs, communications, cameras). It aims to facilitate soldiers in managing multiple drones and autonomous vehicles with simple command prompts like: “Go to this waypoint and watch for enemy forces.”
However, building this software depends on real-world vehicle training. Hence, Foundry, the training zone at a military base where drivers work eight-hour shifts honing the ATVs, gathering data to enhance the model. The base commander even requested Scout’s ATV take turns in security patrols.
Part of Scout’s hypothesis is that VLAs will maximize limited datasets alongside simulation training, delivering a capable driving agent. Though comfortable on trails, the vehicles aren’t fully ready for off-road operations.
Scout is also implementing drones for reconnaissance and as intelligent weapons, leveraging vision language models. They are developing