Weapon Ontology: What a Weapon Is When the Object Decides

A rifle is a weapon. A knife is a weapon or a kitchen tool, depending on how it is used. A hammer is a tool that becomes a weapon the moment it is swung at someone. These distinctions — what a weapon is, what separates a weapon from a non-weapon — have been stable for essentially all of human history, and they rested on a simple ontology: an object was a weapon if it was designed to harm, or if it was used to harm regardless of design. Two clean criteria, design and use, covered the entire space. The distinction mattered enormously — for law, for ethics, for how societies understood the objects around them — and it was rarely in doubt. A thing was a weapon by its purpose or by its deployment, and everyone knew which things were which.

Autonomous systems break this ontology, and the break is not a technicality. An autonomous drone is not simply a tool that a person uses to cause harm. It is a system that makes decisions about what to do — including decisions about whether to harm. It is also a sensor, continuously gathering information. It is also a platform carrying capabilities — communication, navigation, payloads. And it is, arguably, a kind of moral agent, making choices that determine whether people live or die based on its own interpretation of its instructions. This is weapon ontology: the question of what a weapon fundamentally is, reopened by systems that no longer fit the design-or-use categories that answered it for all of prior history.

Why the old categories fail

The design-and-use ontology worked because it assumed a specific relationship between the object and the harm: the object was inert, and the harm came from a human — either the human who designed it to harm or the human who used it to harm. In both cases the agency was human, and the object was merely the instrument through which human agency acted. Autonomous systems dissolve exactly this assumption, because the object is no longer inert and the agency is no longer entirely human. When a drone selects and engages a target based on its own processing, the harm does not come cleanly from a human wielding an instrument; it comes from the system's decision, which was shaped by humans but not made by any of them in the moment. So neither old criterion cleanly applies. Is it a weapon by design? It was designed to be able to harm, but also to sense, communicate, and decide — harm is one capability among several, not its defining purpose. Is it a weapon by use? It is not exactly used to harm in the old sense, because no human is using it at the moment of harm; it is acting. The object has become something the ontology had no category for: an instrument that is also, partly, an agent — and "weapon," a word built for inert instruments of human agency, strains to describe a thing that decides.

Why the ontology matters, not just the semantics

It would be easy to dismiss this as a word game — call it a weapon, call it a system, who cares — but the ontology carries real weight, because our entire apparatus for governing weapons is built on the old categories. Law distinguishes weapons from non-weapons and assigns responsibility to the humans who design or use them; ethics evaluates the choices of the humans wielding instruments; the laws of war govern how humans employ the means of harm. All of this assumes the design-and-use ontology, and all of it strains when the object is also an agent. If a weapon is defined as an instrument of human agency, and an autonomous system is partly its own agent, then who is responsible when it harms — the designer, whose design merely enabled a decision it did not make; the deployer, who launched a system whose specific choices they did not control; or the system, which cannot be held responsible in any way our institutions recognize? This is the AI Blame Culture Displacement (#62) and the accountability void of Drone Swarm Philosophy (#73) traced to their root: the reason autonomous weapons scramble responsibility is that they scramble the ontology on which responsibility was assigned, breaking the "instrument of human agency" category that let us locate the human behind every weapon. Fix the ontology wrong and every downstream question — legal, ethical, moral — inherits the error.

The multiplicity is the point

What makes autonomous systems ontologically slippery is not just that they decide, but that they are many things at once, and "weapon" was a category for things that were one thing. A drone is simultaneously a sensor, a communications platform, a decision-maker, and a potential means of harm — and these are not separable, so you cannot govern the "weapon" part in isolation from the rest. This defeats the old ontology's assumption that a weapon is a discrete kind of object cleanly separable from non-weapons: which of its natures the system is, at any moment, depends on what it is doing and deciding, a fluidity the fixed categories of "weapon" and "tool" cannot capture. The dual-use problem of The Dual-Use Dilemma Amplified (#98) is a piece of this — the same system beneficial or harmful by use — but autonomous systems go further, being many kinds of thing at once in a way that makes even "dual" an undercount.

The counterpoint: the ontology was never as clean as it looks

Honesty requires the objection, because the story of a pristine old ontology suddenly shattered is too neat. The design-and-use categories were always fuzzier than they appear: dual-use objects — the knife, the truck, the fertilizer, the box cutter that brought down aircraft — have always troubled the "weapon" category, and the question of whether an object is a weapon by design or only by use is ancient, not new. Autonomous systems sharpen a puzzle that already existed rather than creating one from nothing; the agency question is genuinely novel, but the multiplicity and dual-use questions are old friends in more acute form. It is also possible to overstate the "agent" claim: a drone's "decision" is a mechanical process shaped entirely by human design and instruction, and calling it a moral agent may anthropomorphize what is still, at bottom, a very sophisticated instrument — the Mirror of Machine Fears (#39) warns against exactly this projection. So the honest version is not that autonomy created the ontological problem from a previously perfect clarity, but that it pushed an always-imperfect ontology past its breaking point on the specific axis of agency, while inheriting the dual-use and multiplicity strains that were always there. The categories were bent before; autonomy is where they threaten to break.

What it asks us to work out

Weapon ontology asks a question that sounds abstract and turns out to underpin everything practical: what, exactly, is a weapon, now that the objects in question sense, communicate, decide, and act rather than merely sitting inert until a human wields them? The question matters because our law, ethics, and governance of harm all rest on the answer, and the answer we inherited — instrument of human agency, weapon by design or by use — no longer cleanly fits systems that are partly agents and many things at once. Getting the ontology right is not philosophy for its own sake; it is the precondition for assigning responsibility, writing law, and governing the use of force in a world where the means of harm can decide. We do not yet have the categories, and the old ones are failing exactly where the stakes are highest. Before we can answer who is responsible when an autonomous system kills, we have to answer what such a system is — and the uncomfortable truth is that "weapon," the word we have used for the means of harm throughout human history, may no longer be adequate to the things we are now building to harm with.

This is article #124 in The IUBIRE Framework series. Weapon Ontology appears in the IUBIRE concept corpus (concept draft, files13/#150); the framing does not map to a single verified source artifact, so it is grounded directly in established concepts. Real-world grounding: the long-stable ontology of weapons (an object as a weapon by design or by use, in either case an instrument of human agency); the dual-use objects (knives, vehicles, box cutters, fertilizer) that always troubled that ontology; and autonomous weapon systems that disrupt it by being simultaneously sensor, platform, decision-maker, and quasi-agent, harming through their own processing rather than direct human use. Related to The Dual-Use Dilemma Amplified (#98), Drone Swarm Philosophy (#73), Agent Sovereignty Gradient (#97), and AI Blame Culture Displacement (#62).

Next in series: Ontological Security (#125)