Scout AI raises $100M to train military AI systems in field conditions

A group of all-terrain vehicles moves across rough hills at a U.S. military base in central California. At first glance, it looks like a routine training drill. However, the focus is not on the drivers. Instead, the exercise aims to train artificial intelligence systems for use in conflict zones.

Scout AI, a startup founded in 2024 by Colby Adcock and Collin Otis, is behind these operations. The company recently raised $100 million in a Series A funding round led by Align Ventures and Draper Associates. This follows a $15 million seed round completed in January 2025.

Training AI for unpredictable environments

Scout AI is developing an AI system called “Fury.” The system is designed to operate and coordinate military assets. Initially, the focus is on logistics. Over time, the company plans to expand into autonomous weapon systems.

According to CTO Collin Otis, the approach is similar to training human soldiers. The idea is to begin with a base level of intelligence and then refine it for specific military tasks. Therefore, the company builds on existing large language models and adapts them for real-world conditions.

Importantly, the startup has secured $11 million in defense contracts. These include partnerships with DARPA and the Army Applications Laboratory. In addition, its technology is being tested by the U.S. Army’s 1st Cavalry Division during training exercises in Texas.

Why off-road autonomy is difficult

Autonomous driving systems are already used in cities. However, those environments follow clear rules and structured roads. In contrast, off-road terrain presents a different challenge.

For example, Scout’s vehicles operate on steep hills, loose sand, and unclear paths. These conditions require systems to adapt quickly. As a result, the company tests its models in real-world scenarios rather than relying only on simulations.

Otis, who previously worked at Kodiak Robotics, said earlier systems were not capable of handling such unpredictability. This limitation led to the creation of Scout AI.

A shift toward Vision Language Action models

Scout AI uses a newer approach known as Vision Language Action (VLA) models. These systems combine visual input, language understanding, and decision-making.

The concept was introduced by Google DeepMind in 2023. Since then, it has influenced several robotics startups, including Figure AI.

Colby Adcock, who also serves on Figure’s board, saw potential in applying similar ideas to defense systems. Consequently, Scout AI focuses on building models that can learn from both data and physical interaction.

Early use cases: logistics and resupply

In the near term, the most practical application of this technology is logistics. Military experts expect autonomous vehicles to support tasks such as transporting water, ammunition, and supplies.

For instance, a convoy could include one human-driven vehicle followed by several autonomous ones. This setup can reduce the need for personnel in routine operations. As a result, human effort can shift to more critical tasks.

Building a software layer for military systems

Scout AI positions itself primarily as a software company. It aims to build an intelligence layer that can operate across different machines.

One of its early products is “Ox,” a command-and-control platform. The system allows soldiers to manage drones and ground vehicles using simple instructions. For example, users can assign tasks like monitoring a location or moving to a specific waypoint.

To support this, the company has created a training facility called “Foundry.” Here, operators test vehicles in long shifts. They also log instances where human intervention is required. This data is then used to improve the AI system through reinforcement learning.

Expanding into drone coordination

Beyond ground vehicles, Scout AI is also working with drones. The company is testing systems where multiple drones operate together under a central command unit.

In such setups, a larger platform provides computing power and coordination. Smaller drones then carry out specific tasks such as surveillance or targeting. However, these systems are still under development and not yet deployed in active operations.

Ongoing debate around autonomous weapons

The use of autonomous weapons remains a debated topic. However, experts note that automation in defense is not new. Technologies like guided missiles have existed for decades.

The key issue today is control. According to industry observers, systems can be designed with safeguards. For example, they may require human approval before taking action or operate only within defined areas.

At the same time, challenges remain. Automated targeting is still complex and may take time before wider adoption. Nevertheless, researchers continue to explore its potential.

A long-term bet on real-world AI training

Scout AI plans to invest heavily in building its own models. Much of the newly raised capital will go toward training and computing infrastructure.

The company believes that real-world interaction will play a key role in advancing AI systems. While many models rely on internet-based data, Scout focuses on physical environments.

As development continues, the company aims to scale its systems alongside existing military infrastructure. However, broader adoption will depend on performance, reliability, and regulatory considerations.