Autonomous Weapons Systems

Autonomous Weapons Systems: Law, Ethics, Policy arrived from Cambridge University Press in 2016, the product of a workshop convened at the European University Institute in Florence and edited by a quintet of international and criminal lawyers — Nehal Bhuta at the EUI, Susanne Beck at Hannover, Robin Geiß at Glasgow, Hin-Yan Liu at Copenhagen, and Claus Kreß at Cologne. It landed at a particular moment in the conversation about weapons that select and engage targets without a human pulling the trigger. The Campaign to Stop Killer Robots had launched in 2013. Christof Heyns, then UN Special Rapporteur on extrajudicial, summary or arbitrary executions, had delivered his report on lethal autonomous robotics to the Human Rights Council that same year, calling for a moratorium. The Convention on Certain Conventional Weapons in Geneva had begun a series of informal meetings of experts in 2014 to take up the question. Inside the US Department of Defense, Directive 3000.09 from 2012 was setting out a policy framework for autonomous and semi-autonomous weapons, including the requirement that they allow commanders and operators to exercise appropriate levels of human judgment. Industry was demonstrating prototypes — the Northrop Grumman X-47B was making its first carrier landings, Israel was fielding loitering munitions like the Harpy and Harop, and the Republic of Korea had deployed the Samsung SGR-A1 sentry along the demilitarised zone. The book is the field’s attempt, at that turning point, to organise what a serious legal and ethical response would have to address.

The argument running through the volume is less a single thesis than a shared starting point: autonomous weapons do something genuinely new to law, ethics, and policy, and the work has to be disaggregated. The chapters resist both the framing that these systems are simply more advanced versions of existing weapons, posing nothing law cannot already handle, and the framing that they are a categorically intolerable evolution that warrants only a preemptive ban. Most contributors take the position that a meaningful response requires sorting which problems are old wine in new bottles — proportionality assessments, distinction, weapons reviews under Article 36 of Additional Protocol I — and which are structurally novel, particularly the questions of accountability when no human has been close enough to the killing decision to be properly responsible for it, and of dignity when life-and-death decisions are delegated to a machine. The editors are explicit that the contributors disagree among themselves on prescriptions; what they share is a refusal to treat the matter as either a technical question for engineers or a romantic question for ethicists.

The book is organised into substantive parts after an introduction. The first takes up the conceptual terrain — what an autonomous weapon actually is, how to draw the line between automation, autonomy, and intelligence, and what scenarios policymakers should actually be worrying about. Definitional work runs throughout, including the International Committee of the Red Cross’s framing in terms of the critical functions of weapons systems, the in-the-loop, on-the-loop, and out-of-the-loop typology familiar from US doctrine, and the DoD 3000.09 categorisation. The second part turns to international humanitarian law, examining whether and how the rules of distinction, proportionality, and precaution can be observed by systems that select targets algorithmically. Marco Sassòli’s contribution, often cited, walks through the IHL framework methodically and argues that autonomy is not in principle incompatible with the rules but that current systems fall short and the burden of proof properly sits with developers. Robin Geiß’s chapter on the Martens Clause — the provision in the preamble to the Hague Conventions and Additional Protocol I that bars conduct contrary to the principles of humanity and the dictates of public conscience — gives a careful account of how that clause might or might not constrain autonomous weapons, and concludes that while the Martens Clause is not a hard prohibition, it sets a substantive expectation that algorithmic killing will face.

The third part is ethical and philosophical. Peter Asaro contributes on the jus ad bellum implications — whether removing human soldiers from the casualty equation lowers the political threshold for resort to force, a worry already familiar from the drone literature. Noel Sharkey, the roboticist who has been one of the most consistent voices for prohibition, sets out his case against ceding targeting decisions to machines on the grounds that no system currently or foreseeably available can satisfy the discrimination requirement in cluttered environments and that the loss of the soldier’s moral judgment is itself a harm. The volume engages Ronald Arkin’s well-known proposal for an ethical governor — a software module that would constrain a system to comply with rules of engagement — and treats Arkin’s claim that machines might one day behave more ethically than human soldiers under battlefield stress as a serious proposition, while several contributors push back hard on the assumption that compliance-checking can substitute for the moral judgment a human brings.

The fourth part takes up responsibility and accountability, and is arguably where the book has been most influential. Hin-Yan Liu’s chapter develops the responsibility gap thesis associated originally with Robert Sparrow, examining how the criminal-law concepts of intent, knowledge, and command responsibility map onto situations where a weapon decides to engage. Dan Saxon’s chapter on individual criminal responsibility under international criminal law works through the corresponding questions for the prosecutor: who would be charged when an autonomous weapon kills civilians in violation of IHL? The programmer? The commanding officer? The manufacturer? The state? The book does not solve this; what it does is map the terrain with enough precision that subsequent work — at the CCW Group of Governmental Experts, in the academic literature, and in domestic legal scholarship — has had a vocabulary and a set of distinctions to use. Susanne Beck’s chapter on the criminal-law perspective is similarly foundational, drawing on continental traditions to ask whether existing doctrines can stretch to cover the new situations or whether new categories of responsibility will have to be created.

A further section on policy and regulation examines the available instruments. Contributors discuss the prospects of a CCW protocol, the precedent of the 1995 Protocol IV on blinding laser weapons — an actual preemptive prohibition that the international community managed to negotiate before the weapons were widely deployed — the role of Article 36 weapons reviews, and the possibility of softer instruments such as political declarations, codes of conduct, and transparency measures. The book is realistic about the limits of what diplomacy in Geneva is likely to deliver. Several contributors note that the major state actors — the United States, Russia, China, the United Kingdom, Israel — have strong reasons to keep their options open, and that any binding instrument will need to find language that the militarily significant states can accept. The editors do not pretend to know what that language is.

The examples threaded through the volume keep the discussion concrete. The Phalanx Close-In Weapon System and the Aegis combat system anchor the discussion of defensive autonomy — both are designed to engage incoming threats faster than a human could react, and both have been operating for decades. The Iron Dome interceptor, then proving itself over Israeli airspace, plays a similar role. The Harpy and Harop loitering munitions, designed to detect and strike enemy radar emitters, are discussed as systems that already operate with substantial autonomy in target selection, raising the question of why they have generated comparatively little controversy. The X-47B carrier-based drone, the Taranis demonstrator in the United Kingdom, and the Russian and Chinese unmanned combat air vehicles under development illustrate where the technology is going. The SGR-A1 sentry comes up repeatedly as a marker of how far autonomy in lethal targeting has already moved in deployed systems. Several chapters reach back to the Patriot missile fratricide incidents of 2003 in Iraq — the system’s autonomous target classification mistakenly engaging friendly aircraft — as a case study in what happens when human operators are encouraged to defer to an automated decision. Christof Heyns’s chapter on the human rights frame applies the same scrutiny to peacetime law enforcement use, where the standards under international human rights law are stricter still than under IHL.

Reception of the volume has been substantial, even when individual chapters have been contested. It became a standard citation in the literature that followed — alongside Heather Roff’s work on responsibility, Paul Scharre’s later book Army of None from 2018, and the policy work coming out of UNIDIR, SIPRI, and the International Committee for Robot Arms Control. The CCW process, which moved from informal experts meetings to a formal Group of Governmental Experts in 2017, drew on the conceptual machinery the book had helped consolidate. The meaningful human control framing, originally promoted by the NGO Article 36, was already in play when the book went to press and has since become the de facto centre of gravity of the diplomatic conversation; the volume’s chapters on the conditions for genuine human judgment fed directly into that framing. Counterpoints exist. Michael Schmitt and Jeffrey Thurnher, whose 2013 Harvard National Security Journal article Out of the Loop argued that autonomous weapons are not per se unlawful under IHL, represent the regulationist position the book treats seriously without endorsing. The dignitarian arguments advanced by some contributors — that being killed by a machine is intrinsically a violation of human dignity regardless of whether the targeting was accurate — have been contested as resting on intuitions that do not survive close scrutiny, by Bradley Strawser and others. The volume’s relative emphasis on European legal traditions has also drawn comment; the criminal-law chapters lean on doctrines that translate imperfectly to common-law jurisdictions, and the volume sometimes underweights US doctrinal developments.

For a reader working through the field on AI in war, the book sits as the legal and philosophical anchor of a small shelf. It pairs with Heather Roff and Richard Moyes’s policy briefs on meaningful human control, with Scharre’s reportorial Army of None for the operational picture, with Armin Krishnan’s earlier Killer Robots: Legality and Ethicality of Autonomous Weapons from 2009 for the longer view, and with the ICRC’s own reports for the humanitarian-law institutional position. What it does not give the reader is much on the AI side itself — the engineering of computer vision, the limits of current machine learning, the failure modes of deep networks in adversarial conditions. Those are subjects the book treats as inputs rather than studies, which is reasonable for a law and ethics volume but means a reader will need to look elsewhere for the technical picture. It also predates the generative-AI inflection of 2022 and after, and the discussion of autonomy in the book is framed largely around classical robotics and rule-based systems rather than learning-based models whose behaviour cannot be fully specified in advance.

What endures from the book is the structural work — the mapping of the questions, the precision about where existing law does and does not bite, the careful presentation of the responsibility-gap problem, and the refusal of easy answers. What dates is the technology census; the named systems have moved on, the X-47B programme was wound down, newer loitering munitions have appeared in the Nagorno-Karabakh and Ukraine conflicts that the book could not have anticipated, and the policy debate has moved from whether to talk about autonomous weapons in Geneva to whether the talks can produce anything binding. A decade on, the chapters on definitions are partly superseded; the chapters on the Martens Clause, on individual criminal responsibility, on the conditions for meaningful human control, and on the strain that autonomous targeting places on the distinction and proportionality requirements are still where serious students of the field begin.