Scout AI Raises $100M for Vertically-Defense Models Trained With Special Ops Veterans

By Jaspal Singh April 29, 2026 Updated: April 29, 2026

Scout AI, the defense-AI startup founded by Coby Adcock, has raised $100 million to train models specifically for warfare applications. The round is unusually large for a Series A and reflects the broader investor scramble for "vertically defense" AI companies — a category that didn't really exist 18 months ago and now has at least eight well-funded entrants.

The pitch is targeted. Scout AI doesn't try to compete with frontier labs on general capability. Instead, the company trains compact models on classified-quality datasets — terrain, sensor feeds, communication intercepts, prior operations — that public-frontier models can never legally see. The result: smaller models that are systematically better at narrow defense tasks than GPT-5 or Claude despite being orders of magnitude smaller.

Why "vertically defense" is a real category now

Two years ago the only path for defense AI was government-prime contractors retrofitting commercial models. That worked badly. Lockheed and Raytheon weren't model-builders, and OpenAI/Anthropic wouldn't license their models for warfare use.

The vacuum produced a new category: companies that build defense-specific models from scratch, on classified data, for government customers exclusively. Scout AI joins Anduril (Lattice software), Shield AI (Hivemind autonomy), Palantir's AIP, and a small number of stealth startups. Investor consensus is that this category has $50B+ of TAM by 2030 even on conservative assumptions — driven mostly by the Pentagon's stated intent to spend $50B+ on AI by 2028.

What Scout's "warfare bootcamp" looks like

The TechCrunch profile that broke the news described an unusual training environment: a converted warehouse in Texas where ex-Special Operations veterans run live red-team exercises against Scout's models. The veterans simulate adversary tactics; the models adapt; the cycle iterates daily. It's reinforcement-learning-from-human-feedback, but the humans are mission-experienced operators rather than data labelers.

That bootcamp model is genuinely novel. Standard RLHF uses crowdworkers; defense RLHF requires people who can credibly simulate an adversary. The talent supply is the actual moat — there are only a few thousand people in the country who can do this work, and Scout is hiring aggressively from JSOC and CIA Special Activities.

The customer dynamics

Scout's primary customer is reportedly DARPA's TIDES program, which funds early-stage operational AI experiments. From there the customer pipeline runs through the Defense Innovation Unit (early operational testing), then to specific service branches (Navy and Air Force first, Army later), then to allied nations (UK, Australia, Israel are the disclosed targets).

The interesting dynamic is that Scout's contract structure includes a "rate card" for outputs that allies can purchase from US-based deployment. That's how Scout monetizes: not selling weights to foreign governments but selling per-query inference under US-supervised infrastructure. That arrangement is increasingly the template for defense AI export.

My Take

Scout AI is one of maybe four companies that have a genuine chance to become category-defining in defense AI. The model is right — vertically integrated, classified-data-trained, ops-veteran-RLHF — and the customer pipeline is solid. The risk is execution. Defense procurement timelines are punishing, and Scout's $100M of runway burns fast at the engineer salaries this kind of work demands. The company also faces the non-trivial possibility that the Pentagon decides to favor in-house government-built models or longstanding prime contractors over startups when actual deployment scale is on the table. I think Scout makes it past Series B but the second-half-decade picture is harder to call. Still, this is the right bet to be making right now.

FAQ

Who led the round? Per TechCrunch, the round was led by a defense-focused fund with participation from Founders Fund and a16z American Dynamism. Specific lead not disclosed.

What's Coby Adcock's background? Former software engineer at Anduril and Palmer Luckey ally; previously founded a smaller defense ML startup acquired in 2023.

Does Scout sell to allies? Yes — through US-hosted inference rather than weight licensing. UK, Australia, Israel disclosed as target customers.

The Bottom Line

Scout AI raises $100M to build vertically-defense AI models trained on classified data with ex-Special Operations RLHF. The defense-AI category is now a real venture-fundable market with clear category leaders forming.