Defining Autonomous Weapon Systems: A schematic framework for refined purpose-driven definitions

Björn Lundgren, Nathan G. WoodarXiv 2026 · arXiv

TL;DR

Instead of pursuing a single, elusive "correct" definition for Autonomous Weapon Systems (AWS), this paper proposes a pragmatic, modular framework. By starting with a simple base definition and adding "plug-and-play" components (like lethality, target class, or domain), stakeholders can create purpose-built definitions that facilitate clearer legal, ethical, and technical debates.

What Problem Does This Paper Address?

For over a decade, international bodies like the UN Convention on Conventional Weapons (CCW) have struggled to regulate AWS, largely because no one can agree on what they actually are. Existing definitions generally fall into two problematic camps:

  1. The Standard View: Often used by the US DoD and ICRC, it focuses on autonomy in "critical functions" (selecting/engaging targets). The authors argue this is too broad—it accidentally includes landmines, trained combat dogs, and even human soldiers under certain interpretations.
  2. The Demanding View: Proposed by some scholars, this requires systems to have high-level agency or "self-learning" capabilities. The authors argue this is too narrow—it relegates AWS to the realm of science fiction and can be "prejudicial" by defining systems in a way that makes them inherently illegal under international law before the debate even begins.

The result is a "talking past one another" problem that stalls policy and regulation.

Methods

The authors employ a pragmatic, engineering-oriented approach to conceptual analysis. Rather than trying to "carve nature at its joints" (a scientific definition), they aim for a definition that is "useful" for solving specific problems.

Their method involves:

  • Critique: Identifying the extensional failures (what a definition includes/excludes) of current views.
  • Desiderata Setting: Establishing that a "good" base definition must be accurate, non-prejudicial, time-neutral (not dependent on whether the tech exists yet), and use-neutral.
  • Synthesis: Constructing a "Base Definition" and a "Modular Schema" that allows users to layer specific characteristics onto that base depending on their needs (e.g., a lawyer needs different specifics than a software engineer).

Datasets

Not discussed in the paper. This is a theoretical and conceptual framework paper within the fields of philosophy, law, and AI governance.

Key Insights and Ideas

  • The "Properly Functioning" Clause: A major insight is that definitions often assume a system is working. The authors argue that a broken or malfunctioning AWS is still an AWS. Their base definition includes the phrase "if properly functioning" to ensure that temporary limitations don't remove a system from regulatory oversight.
  • Machine-Based vs. Organic: To avoid the "biological trap" (where a human or a bat-bomb could be an AWS), the framework explicitly anchors AWS as "machine-based systems."
  • Neutrality as a Virtue: The authors argue against definitions that require "online learning." Why? Because current international law requires weapons to be predictable. If you define AWS as "unpredictable learners," you've effectively banned them by definition, which prevents a nuanced discussion about systems that don't learn in real-time but are still autonomous.
  • The F2T2EA Integration: The paper suggests using the military "kill chain" model (Find, Fix, Track, Target, Engage, Assess) as a modular component. A system might be autonomous in "finding" but not "engaging," and the definition should be able to reflect that granularity.

Key Results

The paper provides a concrete Base Definition:

Autonomous Weapon Systems: Machine-based weapon systems that, if properly functioning, once activated, can identify, select, and engage targets without contemporaneous human input.

It then provides a catalog of Refinement Modules that can be "plugged in":

  • Autonomy Levels: Specifying "Level $n$" autonomy.
  • Target Class: Distinguishing between anti-personnel and anti-materiel.
  • Lethality: Distinguishing between lethal and "less-lethal" designs.
  • Domain-Specifics: Adding modules for Electronic Warfare (EW), Cyber, or Cognitive/Information warfare.

This allows for a "Refined Definition," such as: "A machine-based anti-materiel weapon system with level 4 autonomy in the 'track' and 'target' functions..."

Limitations and Future Work

  • Non-Exhaustive Modules: The authors acknowledge their list of modules (Cyber, Swarming, etc.) is not complete and will need to evolve as technology changes.
  • Stakeholder Honesty: The framework relies on stakeholders being transparent about their political and technical aims. If a state uses a definition specifically to hide its capabilities, no framework can fully prevent that.
  • Implementation: The paper provides the schema, but the actual "group exercise" of stakeholders sitting down to use this framework in a regulatory setting (like the UN) remains a future challenge.

Summary updated

Abstract

The literature on autonomous weapon systems (AWS) includes many proposed definitions for these (types of) systems, with core disagreements between scholars, governments, and international regulatory bodies. In this article, we present and critique prevailing (classes of) definitions, arguing that what is needed is not a singular definition, but rather a definitional framework which allows one to classify different sorts of AWS within the same conceptual family. Recognizing we cannot foresee all future AWS, we show how variations on a broadly accepted definition can be used to produce a definitional schema which can be easily modified in order to classify various concepts belonging to the class of AWS.
BibTeX
@misc{lundgren2026defining,
  author = {Björn Lundgren and Nathan G. Wood},
  title = {{Defining Autonomous Weapon Systems: A schematic framework for refined purpose-driven definitions}},
  year = {2026},
  eprint = {2611.02916},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.02916},
}
Figure