What AI Models for War Actually Look Like

Abstraction: Military-specialized AI startup building models for mission planning and decision dominance

Key points:

• Smack Technologies ($32M funding round) builds AI models trained specifically for military operations via reinforcement learning on war game scenarios with expert analyst feedback — similar to AlphaGo's training approach
• CEO Andy Markoff (ex-Marine Special Operations commander) argues general LLMs like Claude are not capable of target identification and are ill-suited for controlling physical hardware
• Anthropic was declared a DoD "supply chain risk" by defense secretary Pete Hegseth after contract dispute over limits on autonomous weapons use (~$200M contract)
• Academic research found LLMs tended to escalate nuclear conflicts in war game simulations (King's College London, 2025)
• Future use case: automating mission planning drudgery (still done manually with whiteboards); Smack argues AI decision dominance needed against near-peer adversaries like Russia/China

Connections: Anthropic · Smack Technologies · Department Of Defense · Military AI · Autonomous Weapons · AI Safety · Reinforcement Learning

Source: https://www.wired.com/story/ai-model-military-use-smack-technologies/