Four‑seater all‑terrain vehicles are being tested on a military base in central California as part of an effort to train artificial‑intelligence models for use in combat zones. The autonomous ATVs are operated by Scout AI, a defense‑focused startup founded in 2024 by Coby Adcock and Collin Otis. The company announced a $100 million Series A financing round, led by Align Ventures and Draper Associates, following a $15 million seed round in January 2025.
Scout AI is developing an AI system it calls “Fury” to command and control military assets, beginning with logistical support and progressing toward autonomous weapons. According to CTO Collin Otis, the model builds on large language models (LLMs) and is intended to achieve a level of “military AGI” that can be taught to operate complex equipment in the field.
The startup has secured $11 million in contracts from Department of Defense customers, including DARPA and the Army Applications Laboratory. Its technology is among those used by the U.S. Army’s 1st Cavalry Division during training at Fort Hood, Texas, with the expectation that successful systems will be deployed in a 2027 operational cycle.
Training takes place on a hilly, off‑road proving ground at the undisclosed base. Former soldiers on Scout’s operations team run the ATVs through simulated missions, generating data for reinforcement‑learning loops that improve the model’s decision‑making. The company says that vision‑language‑action (VLA) models—derived from recent advances in multimodal LLMs—allow the system to interpret visual cues and execute actions with minimal labeled data.
Scout’s first commercial product is expected to be “Ox,” a command‑and‑control software suite that runs on hardened hardware and enables soldiers to issue high‑level commands to multiple drones and ground vehicles. To refine Ox, Scout’s “Foundry” training range captures driver interventions and feeds them back into the AI pipeline.
Beyond logistics, Scout is experimenting with autonomous drones for reconnaissance and strike missions. The approach involves a larger “quarterback” platform that provides compute resources for a swarm of munition‑carrying drones. The company emphasizes that such weapons can be programmed with geographic and human‑approval constraints to mitigate unintended engagement.
Industry observers note that the technology is still in experimental stages, but its integration into U.S. Army training suggests growing confidence in autonomous ground and aerial systems. Scout’s reliance on existing hyperscaler LLMs for its foundation model, combined with plans to develop a proprietary model, reflects a broader trend of defense firms seeking tighter control over AI capabilities.
The development underscores the Pentagon’s push to field autonomous systems that reduce personnel burden and enhance operational flexibility. As the technology matures, further testing and contracts are expected, positioning Scout AI as a notable player in the emerging military autonomy market.
