Two policy experts have published an article proposing a three-tiered framework for governing the use of artificial intelligence in military and conflict settings. Moira Whelan and Jesper Frant of the National Democratic Institute authored the proposal, arguing for a regulatory structure that moves beyond the current international focus on Lethal Autonomous Weapon Systems (LAWS).
According to the authors, current discussions at the United Nations Convention on Certain Conventional Weapons (CCW) are too narrowly concentrated on fully autonomous weapons. Their article states that this focus overlooks the immediate ways AI is already being integrated into military operations. They identify existing applications in surveillance, data analysis for targeting, and the dissemination of disinformation as areas requiring clearer governance.
The framework detailed by Whelan and Frant organizes AI applications into three distinct categories based on risk:
Tier 1: Prohibited Uses. This category would include AI systems that are fundamentally inconsistent with international humanitarian law and democratic principles. The authors suggest this would encompass applications such as AI-driven social scoring systems used to determine combatant status or systems incapable of adhering to the principles of distinction and proportionality in conflict.
Tier 2: Restricted Uses. This tier covers high-risk AI applications that would require strict regulation, testing, and meaningful human oversight. Systems used for intelligence gathering, identifying targets, or cyber operations would fall into this category, according to the proposal. The authors write that such systems would necessitate rigorous protocols to ensure human control and accountability.
Tier 3: Permitted Uses. The lowest-risk category would include AI applications that support administrative or logistical functions with standard oversight. The article provides examples such as AI used for predictive maintenance on military vehicles or for optimizing supply chain logistics. These uses, the authors state, do not directly involve the application of force and carry a lower risk profile.
Whelan and Frant write that their proposed framework is intended to provide a more practical and comprehensive approach for policymakers. They assert that such a structure would enable governments to guide the development, procurement, and deployment of military AI technologies while establishing clearer international norms. The authors position the framework as a tool to advance U.S. and allied interests in responsible AI adoption.
