What AI Models for War Actually Look Like

While companies like Anthropic impose restrictions on military use of their AI, a new startup called Smack Technologies is building advanced models specifically for warfare. Founded by former Marine special forces commander Andy Markoff and Tinder's ex-VP of technology Dan Gould, the company just announced a $32 million funding round to develop AI that learns optimal mission plans through trial-and-error war game scenarios, with expert analysts providing feedback signals. Markoff argues general-purpose LLMs like Claude are ill-suited for military applications as they lack training on military data and human-level understanding of the physical world needed for tasks like mission planning or controlling hardware.

Smack's approach represents a growing sector of specialized defense AI, emerging amid controversy over Silicon Valley's role in military tech. The funding follows a breakdown in talks between Anthropic and the Department of Defense over a $200M contract, where Anthropic sought to limit autonomous weapons use—a restriction Smack appears to avoid. The startup aims to automate the drudgery of manual mission planning (still done with whiteboards) to give the US 'decision dominance' against near-peer adversaries. However, experts warn clearer red lines are needed, especially after research showed LLMs tend to escalate nuclear conflicts in simulations, raising critical questions about reliable deployment in high-stakes scenarios.