The Autonomous Weapons Debate Still Has a Definition Problem

Cfr | Michael C. Horowitz, Lauren Kahn

Semantic drift regarding autonomous weapons creates a persistent 180-degree problem among policymakers and negotiators who treat autonomy as a simple binary rather than a technical spectrum. Modern military forces have integrated varying degrees of automated targeting and guidance into platforms for decades, ranging from early historical systems to advanced capabilities like the Patriot missile system and shipborne Close In Weapon Systems.

This persistent ambiguity now directly affects legislative oversight, as seen in the Senate’s draft National Defense Authorization Act, which contains dozens of pages of micro-level regulations for Pentagon AI use without fully distinguishing operational decision-making layers. Operational targeting involves distinct phases, separating the initial political decision to conduct a military operation from the subsequent operational targeting process and the eventual deployment of specific weapon effectors. To resolve these ongoing classification challenges, analysts can evaluate systems across three fundamental operational criteria: whether a machine selects the target, whether it engages the target, and whether human supervision allows for real-time monitoring and intervention.

Comment

Public debates surrounding autonomous weapon systems routinely conflate distinct technical tiers, misinterpreting narrow edge-deployed computer vision models for frontier general-purpose platforms like large language models. This semantic confusion directly shapes legislative mechanisms such as the Senate's draft National Defense Authorization Act, imposing rigid compliance frameworks that fail to account for decades of entrenched automation within legacy architectures like the Close In Weapon Systems or the Patriot missile battery. Consequently, regulatory proposals risk inadvertently penalising foundational force-protection systems that have operated with minimal human intervention during terminal engagement phases for generations.

Strategic Question for Discussion

How does the conflation of narrow edge-deployed artificial intelligence with general-purpose large language models distort the regulatory framework within the draft National Defense Authorization Act for existing terminal defense platforms?

The available evidence indicates that treating tactical computer vision models as equivalent to frontier large language models mischaracterises the computational constraints of fielded weapon systems. This analytical mismatch risks imposing compliance burdens on legacy architectures that operate on deterministic, well-bounded subroutines rather than stochastic reasoning.

Share your assessment in the comments below.

To read in details click here