How Lessons Learned from the CCW Can Help Guide the Regulation of AWS
Autonomous weapon systems have become a familiar set piece in contemporary International Humanitarian Law (IHL) debates. The questions are by now well-rehearsed: does the existing law of armed conflict already regulate these systems; can algorithms reliably distinguish lawful targets; and how do we allocate responsibility when lethal force is mediated through software? These are the right questions, but they are too often answered in the abstract. The more instructive guidance is not abstract, it is historical. It lies in the history of the Convention on Certain Conventional Weapons (CCW), a framework treaty designed to constrain weapons “deemed to be excessively injurious or to have indiscriminate effects,” implemented through protocols that target particular technologies.
If one reads the CCW closely, a pattern emerges. Where States drew clear, technically legible regulatory lines early in weapons development, as with blinding laser weapons, the law shaped development pathways. Where States relied on incremental restriction for a weapon already entrenched in their arsenals, as was the case with landmines, regulation became longer, harder, and ultimately vulnerable to geopolitical reversal. And most importantly, where definitional choices left loopholes, such as the case with incendiary weapons, the law struggled to keep up with how weapons were actually used.
For AWS the central issue is not whether IHL “applies,” it does. The question is whether States will do what arms control has required before: set durable, weapon-specific constraints before battlefield practice hardens into a new baseline. The CCW is not a theory of regulation; it is a record of what has worked, what has failed, and why. If States are serious about limiting foreseeable humanitarian harm from autonomy in critical functions, that record is an important place to start.
CCW Model
The CCW’s architecture is often misunderstood as a grab bag of narrow prohibitions. This is not the case. In reality, the CCW reflects a disciplined legal method of identifying a weapon characteristic that tends toward unlawful effects and then codifying constraints that are operationally meaningful and technically reviewable.
That method matters for AWS because the gravest concern is not abstract. It is the delegation of critical functions to systems that may operate without meaningful human control, particularly in environments where legal judgment is more readily required. This is precisely the kind of “weapon-effect” problem the CCW was designed to address.
So what do the protocols teach us?
Protocol I (Non-Detectable Fragments)
CCW Protocol I prohibits weapons whose “primary effect” is to injure by fragments that escape detection by X-rays. This is arms control at its cleanest. A narrow, technically definable feature is identified as inconsistent with humanitarian principles and the treaty draws a categorical line in the sand.
Two lessons carry over to AWS.
First, prohibitions can be targeted at a design feature. Protocol I does not ban “fragmentation” generally; it bans a specific fragment property that produces a specific humanitarian harm. That same logic supports AWS rules framed around autonomy in critical functions, not broad categories of weapons delivery but the specific delegation of target selection and engagement beyond meaningful human control.
Second, early, precise prohibitions avoid the “it depends” paralysis that often stalls regulation. Protocol I did not wait for widespread battlefield practice to crystallize. It set a standard that states and engineers could understand, test, and comply with.
Protocol II and its 1996 Amendment (Mines, Booby-Traps and Other Devices) and the Ottawa Process
Protocol II (and the 1996 Amended Protocol II) is the CCW’s most direct illustration of what happens when regulation comes late and is limited by bureaucracy. While Amended Protocol II strengthened restrictions by addressing issues like marking, monitoring, and broader rules around mine use, it did not deliver a total ban.
The result is a familiar arms-control story. Because CCW consensus could not reach prohibition, States and civil society moved to an alternative track, the 1997 Anti-Personnel Mine Ban Convention (Ottawa Convention), which entered into force in 1999.
This sequence offers three AWS lessons.
1) Waiting for consensus within the CCW can, as a practical matter, mean accepting the least ambitious outcome that the most reluctant States will tolerate. Amended Protocol II was unquestionably a humanitarian improvement. But its trajectory also illustrates a structural constraint. Where decisions are effectively hostage to unanimity, the CCW can function as a regulatory ceiling as much as a vehicle for progressive development. The turn to the Ottawa process was not a rejection of the CCW’s objectives; it was an acknowledgment that the forum’s decision rules limited what could be achieved within it.
That institutional lesson is difficult to ignore in the AWS context. CCW discussions have generated useful exchanges on human control, weapons reviews, and compliance with IHL, yet the repeated difficulty in translating those discussions into binding regulations shows that a breakaway forum may be required to successfully regulate AWS in a meaningful way, as is already emerging.
2) Post-hoc regulation becomes an endurance contest. Landmines required decades of clearance, victim assistance, and norm-building. And even after the stigma became powerful, strategic anxiety in the current geopolitical landscape has driven renewed reconsideration with multiple European states announcing or pursuing withdrawal plans from the Ottawa Convention in 2025. The key point for AWS is not to relitigate those decisions, it is to recognize how fragile humanitarian gains and regulation can become when a weapon is perceived as militarily “useful” and is already woven into defence planning.
3) If AWS become normalised on the battlefield first, the world may later face an Ottawa-scale cleanup problem, but one without a physical artifact to clean up. Although highly difficult, dangerous and problematic, a minefield can be mapped and cleared. Algorithmic targeting practices diffuse through software, doctrine, and procurement supply chains. Once embedded, it is highly likely they will be significantly harder to “clear” from militaries and arms markets, creating ongoing problems that reach far beyond any successful later-stage regulation.
Protocol III (Incendiary Weapons)
Protocol III restricts incendiary weapons and defines them as weapons “primarily designed” to set fire to objects or cause burn injury. That definitional choice, “primarily designed,” has produced long-running debates about whether certain munitions with incendiary effects fall inside or outside the Protocol’s core restrictions.
The AWS parallel is immediate. If regulation is drafted around labels, the law will chase engineering semantics. Similarly, if regulation is drafted around a single metric, States will argue about gradations while deployment accelerates.
Protocol III’s lesson is that functional, effects-based drafting is more resilient than category-based drafting. For AWS, that means rules should attach to what the system is permitted to do, not what it is called. The most legally defensible approach is to regulate critical functions in terms that map onto IHL compliance requirements: selecting and engaging targets; operating across wide geographic and temporal scopes; and applying force based on generalized target profiles in environments where civilians are foreseeably intermingled.
Protocol IV (Blinding Laser Weapons)
Protocol IV is the arms control precedent AWS advocates cite for good reason. It prohibits employing laser weapons specifically designed to cause permanent blindness, and it also addresses transfer.
It is difficult to overstate how important this example is. Protocol IV shows that States can:
– identify an emerging capability;
– treat its defining effect as unacceptable;
– prohibit it before widespread battlefield normalization;
– and still preserve legitimate non-prohibited uses (range-finding, targeting aids, and other laser functions not designed to blind).
That is the exact shape AWS regulation should take, and is the shape that regulation is taking in the current negotiations. Ban what should never become normal, regulate the rest. The relevant AWS analogue is not a blanket ban on autonomy. It is a prohibition on systems that select and engage human targets without meaningful human control, and strict regulation of other autonomous functions (especially where the system operates in complex civilian environments).
The legal rationale fits the CCW’s own DNA of preventing means of warfare that tend toward indiscriminate effect or unacceptable harm and do so early enough that “normal practice” does not evolve past what is legally acceptable and become the baseline.
Protocol V (Explosive Remnants of War)
Protocol V does not ban a weapon, it tackles the predictable humanitarian harm left behind through obligations to mark, clear, remove, destroy, and cooperate.
The AWS lesson here is important. Even if States cannot yet agree on categorical prohibitions, they can agree on harm-mitigation obligations that make accountability real:
– recording and information-sharing duties (what was used, where, under what parameters);
– requirements that facilitate investigation of civilian harm;
– assistance and remedial measures when autonomous functions cause unlawful outcomes;
– and design expectations (fail-safes, audit logs, and traceability) that make responsibility practicable rather than theoretical.
Protocol V teaches that arms control can regulate not only use but also the life cycle of harm before, during, and after conflict. In AWS terms, that points toward treaty requirements for transparency, testing, and post-incident review mechanisms that are currently inconsistent across States.
The Overarching Lessons: Arms Control Works Best Before Normalization
If we put learnings from the CCW protocols together the historical message is consistent:
1. Early, narrow, technically legible prohibitions (Protocol I; Protocol IV) can reshape development pathways.
2. Incremental restriction for entrenched weapons (Protocol II; Amended Protocol II) may help, but often leads to a later breakout treaty (Ottawa) and remains vulnerable to strategic reversal.
3. Definitions matter as much as prohibitions (Protocol III).
4. Life-cycle obligations can reduce humanitarian harm (Protocol V).
AWS regulation should be designed as an application of these lessons, not as a novel philosophical experiment. Drawing directly from the protocols, AWS regulation should aim to be:
– Function-focused: prohibit or restrict autonomy in target selection and engagement (Protocol I & III lesson).
– Pre-emptive where the effect is unacceptable: ban systems that select and engage human targets without meaningful human control and constrain the transfer of such capabilities (Protocol IV lesson).
– Resilient to definitional gaming: avoid definitional loopholes by anchoring obligations to the system’s permissible operational behaviour, foreseeable environments, and human control requirements (Protocol III lesson).
– Life-cycle accountable: require logging, traceability, information sharing, and remedial measures analogous to post-conflict obligations (Protocol V lesson).
The point is not to relitigate the CCW’s past, it is to use it as a field-tested guide to what works.
Conclusion
The CCW record is not a museum piece; it is a warning about how quickly the normative baseline can shift once a capability becomes operationally ordinary. Protocol I and Protocol IV demonstrate that early, technically intelligible prohibitions can channel weapons development away from effects the law should never tolerate. Protocol II shows that incrementalism, once a weapon is embedded in doctrine and stockpiles, is costly, slow, and politically reversible, often forcing later breakout processes to do the work that could not be done in time. Protocol III shows that drafting choices are not ancillary: definitions become operational loopholes, and loopholes become practice. Protocol V shows that when outright bans are politically unattainable, robust life-cycle obligations can still operationalise accountability and reduce humanitarian harm.
Taken together, these lessons point to the same regulatory imperative for AWS: define the prohibited effect (autonomous selection and engagement of human targets without meaningful human control), regulate the remainder through function-based constraints and transparency obligations that make legal responsibility real, and do so before deployment patterns harden into a new normal. Encouragingly, that is broadly the direction in which regulation is moving. But direction is not destination. If States allow negotiations to drift into definitional ambiguity or non-binding “best practices,” the battlefield will supply the default rules. If the international community waits for battlefield practice to settle the question, it will not merely be late, it will have allowed technical momentum to create our new legal reality.
***
Natalie Nunn is a Senior Legal Consultant with Lexbridge Lawyers, specialising in international humanitarian law (IHL) with a focus on IHL compliance in contemporary conflicts and the regulation of military applications of artificial intelligence.
The views expressed are those of the author, and do not necessarily reflect the official position of the United States Military Academy, Department of the Army, or Department of Defense.
Articles of War is a forum for professionals to share opinions and cultivate ideas. Articles of War does not screen articles to fit a particular editorial agenda, nor endorse or advocate material that is published. Authorship does not indicate affiliation with Articles of War, the Lieber Institute, or the United States Military Academy West Point.
Photo credit: U.S. Air Force, Airman 1st Class Christian Clausen
