The Ungoverned Battlefield: Nine Theatres and the Question No One Answered
[Navigation links not applicable for narrative series]
Nine Fridays ago, this series opened with an autopilot that could not tell a barrage balloon from a Spitfire. A machine given a lethal decision it was not equipped to make, in the earliest days of automated air defence. The series closes in a conference room in Geneva, where the Convention on Certain Conventional Weapons has met, on and off, for thirteen years, to discuss whether to regulate autonomous weapons. It has produced not one binding rule.
During those same thirteen years, autonomous weapons were deployed in Libya, Azerbaijan, Ukraine, Ethiopia, Gaza, Lebanon, and Iran. Seven theatres, on three continents, over the exact span in which the world's governments were meeting to decide whether such weapons needed rules. Every episode in this series documented a different theatre, a different vendor, a different failure of accountability. The pattern held regardless of which state was involved. That is the point of naming all seven: the failure is not theatre-specific. It is universal.
The gap this series has traced, between what these systems can do and what anyone has authorised them to do, is not a structural absence waiting on better diplomacy. It is a preference, expressed by the states with the power to close it, at every meeting, every vote, every drafting session, for thirteen straight years.
The responsibility gap
The problem was named before the weapons existed. In 2007, the philosopher Robert Sparrow published a paper called "Killer Robots," working through a question that seemed abstract at the time and is now operational: who answers when an autonomous weapon kills the wrong person?
Sparrow considered three possible answers, and rejected each. The programmer wrote code that functions within general parameters; the specific lethal outcome is not their intent, and holding a developer responsible for an autonomous system's unforeseen choice, Sparrow argued, is like holding parents responsible for the actions of their adult children. The commander authorised deployment in a defined context, not the specific engagement that followed. And the machine itself cannot be punished, cannot be shamed, has no interests any sanction can touch.
Sparrow called the result a responsibility gap: a situation where the moral responsibility a harm deserves exceeds what can be assigned to any single agent. He argued the gap made autonomous lethal weapons inherently unjust, not because they choose wrongly, but because no one can be justly held to account for what they choose.
Nineteen years on, the gap has not closed. It has scaled.
Eleven years, zero reviews
Consider the one instrument that was supposed to prevent exactly this. The United States policy on autonomous weapons, Directive 3000.09, was signed in 2012. It required a senior review before any autonomous weapons system entered development.
In 2019, a Pentagon spokesperson confirmed that, to that date, no weapon had ever undergone that review. Eleven years, zero reviews. The 2023 revision of the directive quietly removed the word "control" from the directive's own operative definition, for what an official described as technical accuracy. The same revision added a waiver for urgent military need, undefined.
Meanwhile, the standard for what counts as meaningful human control has never been agreed. The International Committee of the Red Cross asks whether a human genuinely understood the situation and could withhold approval. The Campaign to Stop Killer Robots asks whether a human individually identified the target. One national standard asks only whether the review was appropriate to the operational tempo. Under that last standard, the twenty-second review documented in a recent campaign, examined in the previous episode of this series, is not a shortcut around control. It is what control has been defined to mean.
When the definition bends to fit the practice, the practice is no longer being governed. It is being described.
The honest counter-argument
The strongest argument on the other side deserves a direct answer, not a dismissal. It runs like this: military AI may make fewer targeting errors than a frightened human soldier under fire, and a system that kills more precisely satisfies a real humanitarian obligation to reduce civilian harm. That argument is honest. It is not made in bad faith.
It is also not enough, because precision and accountability are different variables. A highly accurate weapon, deployed by the wrong decision, against the wrong target, with no one responsible for the outcome, has not closed the gap. It has dressed it up. Precision is a property of the weapon. Accountability is a property of the system that surrounds it. The two do not substitute for each other.
In March 2026, a court heard exactly this argument in miniature. An artificial intelligence company had refused to allow its model to be used for fully autonomous weapons, and a government branded that company a national security threat for declining. The presiding judge called the government's position Orwellian, and granted a preliminary injunction. The line held, that time. Whether it holds next time was never the right question. The right question is why a courtroom, rather than a treaty, was the mechanism that finally worked. When the only functioning check on autonomous lethality is a single judge responding to a single corporate refusal, the governance architecture has already failed. A court is not a substitute for a rule.
When no one is counting
The gap does not stay at the scale of a single strike. It widens as the systems scale, and the systems are scaling fast.
In January 2026, a demonstration showed a large number of drones launching from a single vehicle, dividing their own targets among themselves, and striking without a human individually choosing which one died. The one-to-many command structure is no longer experimental. When one operator commands dozens of platforms simultaneously, and the platforms allocate lethal decisions among themselves in real time, Sparrow's responsibility gap does not just survive the scale-up. It becomes unanswerable. There is no single human decision to review, because no single human made one.
No treaty sets a human-review standard for a system of that kind. No court has tested who answers when the decision that killed someone was never made by any identifiable person. The oversight mechanisms that exist were built for a different architecture: one weapon, one operator, one decision, one review. The architecture being fielded now has dissolved every one of those singulars into a distributed process, and the governance has not followed.
Article 36 and the weapons review that never happened
Step back to the procurement and sovereignty question, because this reaches every organisation that builds, integrates, or advises on these systems, not only the states that field them. Article 36 of the 1977 Additional Protocol to the Geneva Conventions requires that any state studying, developing, or acquiring a new weapon determine whether its use would, in some or all circumstances, be prohibited by international law. It is the one legal instrument that speaks directly to a weapon before it is deployed rather than after. Applied honestly to a swarm that reallocates lethal decisions across dozens of platforms in real time, an Article 36 review cannot be satisfied, because there is no single use, no single circumstance, and no single decision to assess. Export licences and end-use assurances were written for one weapon transferred to one named operator. For enterprises inside allied supply chains, including in New Zealand, the practical question is whether your governance can name who is accountable for a system you supplied, integrated, or advised on, when it makes a decision no human individually authorised.
AI-first by executive decision
The largest shift in how militaries select targets since precision-guided munitions arrived was not legislated. It was announced. A strategy memorandum directed one military to become an artificial-intelligence-first force, cleared internal safeguards it described as blockers, and wrote into every new contract a clause stating that a supplier's only permitted objection to a use of its technology was whether that use was lawful. Not ethical. Not safe. Lawful, as determined by the buyer.
A sitting legislator said it plainly in committee testimony in March 2026: the legislature had not set any statutory framework for how artificial intelligence may be used in lethal military operations. Nothing followed the statement. No hearing. No vote. No bill.
For the civilian on the receiving end, the position is worse than unregulated. An international criminal court has jurisdiction in principle, but the states running the largest autonomous weapons programmes are not meaningfully bound by it. Domestic courts have consistently declined to hear these cases at all, even where the person killed held citizenship of the country that killed them. That failure holds regardless of whether the state involved is an ally or an adversary. No treaty. No court. No vote. The oversight that exists is a reporting requirement, not an authorisation requirement, and a report is not consent.
The five conditions
So the series closes where it must: with a test. Not a prediction, not a warning, but a falsifiable standard that anyone can check.
Five conditions would count as real evidence that the accountability gap is closing. A binding treaty with verification and enforcement. An operational human-control standard that could be independently checked. Parliamentary oversight with access to the classified detail. Judicial review of an actual autonomous targeting decision. And a demonstrated record of at least one person actually being held accountable for an autonomous system's lethal error.
As of the writing of the book this series is drawn from, not one of the five has been met. Not one binding treaty. Not one agreed and checkable control standard. Not one legislature with real oversight access. Not one court that has reviewed an autonomous targeting decision on its merits. And not one case, in any jurisdiction, where a person was held legally accountable for an autonomous system's lethal error.
A majority of the public, surveyed across twenty-eight countries, opposes fully autonomous weapons outright. That opposition has no institutional channel through which to become policy. The people most affected were never asked, and the mechanisms that would let them answer were never built.
This series began with a weapon that could not tell a barrage balloon from a Spitfire. It closes with weapons that can select a human being for death faster than any court, legislature, or treaty has been asked to keep pace with. The gap was never a mystery, and it was never an accident. It was a series of choices, made by people with names, in institutions with addresses, none of whom were ever required to ask permission first. The engineers chose the training data. The commanders chose the tempo. The diplomats chose, thirteen times over thirteen years, not to agree a rule. And the publics in whose name all of it was done were never given a vote.
The accountability gap is not waiting on better technology or better diplomacy. It is waiting on a decision that the people with the power to make it have, so far, chosen not to make. That choice is still open. The first step toward closing the gap is naming it accurately: not as an absence, but as a preference. What would it take for someone, finally, to be asked?
The views expressed in this article are entirely my own, informed by more than 30 years of professional experience in architecture, security, and technology leadership in New Zealand. They do not represent the views of my employer, any government agency, or the New Zealand government. My commentary on legislation and policy is analytical, drawing on publicly available sources and my professional expertise in architecture, security, and AI governance. I follow the Public Service Commissioner's Code of Conduct for the Public Sector and social media guidance.
Andreas Hamberger is a Wellington-based enterprise architect and governance practitioner with more than 30 years of experience across New Zealand's public and private sectors. He holds TOGAF, IAPP, and AMInstD credentials, is an Associate Member of the Institute of Directors New Zealand, and is the founder of Te Pono Limited. Lethal by Design: Military AI and the Crisis of Human Control examines the accountability gap in autonomous and AI-assisted weapons, and the choices that built it.
I use AI tools, including Sudowrite, Claude, Perplexity AI, DeepSeek AI, ChatGPT, Grok, Copilot, Openart and Gemini, as deliberate production tools, not ghostwriters. This is consistent with my position: AI amplifies human judgement; it does not replace it. The frameworks, arguments, and editorial decisions in this series are original work. AI accelerated the process. The thinking is mine.
[1] Lethal by Design: Military AI and the Crisis of Human Control, Chapters 22-27. Te Pono Limited, 2026.
[2] Sparrow, Robert. "Killer Robots." Journal of Applied Philosophy 24, no. 1 (2007): 62-77. https://doi.org/10.1111/j.1468-5930.2007.00346.x
[3] United Nations Office for Disarmament Affairs. "Group of Governmental Experts on Lethal Autonomous Weapons Systems: Session Records 2014-2026." https://www.un.org/disarmament
[4] United States Department of Defense. "Directive 3000.09: Autonomy in Weapon Systems." 21 November 2012; revised 25 January 2023. https://www.esd.whs.mil
[5] Congressional Research Service. "International Discussions Concerning Lethal Autonomous Weapon Systems." 2019-2024. https://crsreports.congress.gov
[6] International Committee of the Red Cross. "ICRC Position on Autonomous Weapon Systems." 12 May 2021. https://www.icrc.org
[7] United States District Court, Northern District of California. Preliminary injunction order, March 2026. Court docket record.
[8] DefenseScoop. "One-to-Many Drone Swarm Demonstration at Camp Blanding." January 2026. https://defensescoop.com
[9] United States Senate Armed Services Committee. Hearing transcript, 5 March 2026. https://www.armed-services.senate.gov
[10] Ipsos. "Global Survey on Autonomous Weapons: 28 Countries." 2021. https://www.ipsos.com
[11] International Committee of the Red Cross. "What Does IHL Say About Autonomous Weapons?" Geneva, 2021. https://www.icrc.org
[12] Additional Protocol I to the Geneva Conventions, Article 36 (Weapons Review Obligation). 1977. https://ihl-databases.icrc.org