Stop Using AI as a Tool, Command It

The U.S. Department of Defense and military leaders must stop treating artificial intelligence merely as a simple tool supporting humans, and instead elevate agents to the status of military subordinates, managing them through mission command principles.

AI Agents Must Be Subordinates

The philosophy of mission command cannot exist without subordinates and a clear intent. Military leaders must stop viewing agents merely as tools that augment human capabilities.

The Department of Defense must build military artificial intelligence forces by organizing agents under commanders, qualifying them for actual missions, and allowing humans to assume the risk.

Mission-Oriented Certification and Understanding

Raw task outcomes and test frameworks should be used to evaluate the capabilities of agents. Success rates and various benchmarks measured by Model Evaluation and Threat Research guide commanders.

While models such as systems developed by Anthropic successfully complete some complex tasks, they may fail at simpler tasks. Commanders must evaluate these results as mission-oriented certifications.

Mission Command Instead of Remote Computing

The mission command approach, based on the principles of centralized planning and decentralized execution, requires the ability to act even when connectivity is severed in conflict environments.

Sending data to vendors' cloud platforms leads to delays and dependencies that enemies can intercept. Commanders must keep models under their own control.

Use Models You Can Afford to Lose

Rather than depending on expensive and massive model architectures, small and open-weight models should be preferred. Models such as the Qwen series developed by Alibaba Cloud can achieve success.

Military units can ensure security and accessibility by locally running kill-switch mechanisms and purpose-fit small-scale models to limit risk.