A Treaty for Autonomous Weapons: What Should Be Prohibited and What Human Control Should Require
- Edmarverson A. Santos
- Jul 10
- 4 min read
Policy Brief · International Law & Emerging Military Technology
This policy brief argues that states should negotiate a legally binding Treaty for Autonomous Weapons built around a two-tier model: prohibiting systems that cannot be lawfully controlled and regulating all remaining autonomous weapon systems through concrete duties on design, testing, authorization, supervision, intervention, traceability, and legal review.
Executive Summary
Autonomous weapon systems are not outside international humanitarian law. Existing rules on distinction, proportionality, precautions in attack, and weapons review already apply to their development and use. The problem is more specific: the current law does not clearly define which autonomous systems should never be developed or used, nor does it specify how much human judgment is legally required before a machine selects and engages targets after activation.
The brief proposes a legally binding treaty rather than another non-binding statement of principles. Its central argument is that regulation should not depend on vague labels such as “AI weapons” or “killer robots.” The treaty should use a functional definition: a weapon system that, after activation, selects and applies force to targets without further human intervention.
The proposed treaty would prohibit two categories. First, systems whose behavior, operation, or effects cannot be sufficiently understood, tested, predicted, explained, and constrained by the humans responsible for their use. Second, systems designed or used to select persons for attack on the basis of generalized target profiles. Other autonomous systems, especially narrowly bounded anti-materiel systems such as certain point-defense systems, would not be banned solely because they operate autonomously. They would remain lawful only if they satisfy strict human-control requirements.
The brief also preserves the existing legal uncertainty. It does not claim that every proposed rule is already binding law. Some unpredictable systems may already be unlawful under IHL if their effects cannot be limited as required. The broader treaty architecture, however, is a reform proposal: it would convert general IHL obligations into testable rules for procurement, deployment, command responsibility, weapons review, and operational use.
Key Recommendations
1. Adopt a functional treaty definition of autonomous weapon systems based on autonomy in target selection and engagement, rather than technology labels, AI terminology, or fixed levels of autonomy.
2. Prohibit autonomous weapon systems that fail a predictability and constraint test, including systems whose effects cannot be reasonably anticipated, limited, tested, or reviewed by responsible human decision-makers.
3. Prohibit autonomous systems that independently select persons for attack through generalized target profiles, while preserving the distinction between autonomous targeting of persons and human-authorized strikes assisted by tracking or homing functions.
4. Codify an integrated human-control standard requiring identifiable human authority, technical and contextual understanding, conscious authorization, predictability, operational limits, intervention capacity where feasible, sufficient time for legal judgment, traceability, and recurring review.
5. Require lifecycle legal review for all covered systems, including review during study and development, before acquisition or deployment, before transfer, and after material changes to software, target profiles, mission parameters, or operating environments.
6. Regulate permitted anti-materiel autonomous systems through narrow limits on target type, geographic area, duration, scale, civilian proximity, supervision, logging, and review.
7. Include national implementation duties covering procurement law, military doctrine, rules of engagement, training, accountability procedures, export controls, and penalties proportionate to the violation.
8. Pursue a dual diplomatic strategy: continue work through the CCW process, but prepare a General Assembly-convened or independent negotiating route if consensus prevents a formal mandate.
Also read
Why a Treaty for Autonomous Weapons Matters
The legal debate on autonomous weapons has moved beyond the basic question of whether IHL applies. It does. The harder question is how rules written for human judgment should operate when a weapon, after activation, can select and engage targets through sensors, software, and predefined or learned processing rules.
That gap matters because legal responsibility cannot be assigned to a machine. Commanders, operators, state officials, and weapons reviewers remain responsible for compliance. If those human actors cannot understand the system’s target-recognition logic, reasonably anticipate its behavior, limit its operating area and duration, or reconstruct how it acted, accountability becomes thin in practice even if it remains formally intact.
The brief’s two-tier structure is useful because it avoids two weak extremes. A total prohibition on all autonomous weapons would likely fail to distinguish between high-risk anti-personnel systems and narrowly bounded defensive systems used against incoming missiles or munitions. A purely permissive approach would treat ordinary IHL standards as sufficient without answering how those standards should be implemented in design, testing, authorization, and deployment.
The proposed treaty would draw a sharper line. Systems that cannot be controlled in a legally meaningful way would be banned. Systems that select people for attack based on generalized profiles would also be banned because they pose distinct risks of distinction, surrender, incapacitation, civilian status, and direct participation in hostilities. Other systems would be restricted, not ignored.
The institutional issue is equally important. The CCW process has produced technical discussion, guiding principles, and a rolling text, but consensus has not yet delivered a negotiating mandate. The brief, therefore, treats the 2026 Review Conference as a political decision point. If states cannot move the issue into formal treaty negotiations within the CCW, supporters of binding regulation should not allow procedural deadlock to become the policy.
The contribution of the brief is practical rather than rhetorical. It sets out what a treaty should prohibit, what human control should require, how weapons review should operate across a system’s lifecycle, and how states could pursue negotiations without pretending that all military powers already agree. It is a legal reform proposal grounded in existing IHL, but it does not confuse proposed treaty obligations with law already in force.
Suggested Citation
Edmarverson A. dos Santos, A Treaty for Autonomous Weapons: What Should Be Prohibited and What Human Control Should Require, Diplomacy & Law, 2026.
