International Treaties on AI Weapons: United Against Turbo-Killer Robots

It only takes about 20 seconds and a few clicks on the computer. This is followed by an explosion in the middle of a residential area. “Lavender” is the name of the AI-supported system with which an officer in the Israeli army can mark a name and make a person a target in under a minute.

Then a second system takes over called “Where’s Daddy?”, in German “Wo ist Papa?”. It specializes in locating people at home. The target, a man, is usually in a residential building. It was right there, in the middle of the neighborhood, with his wife and children, that the next bomb fell. The Israeli-Palestinian online magazine +972 and its sister medium Local Call report on the use of AI systems by the Israeli military. The research is based on statements from Israeli secret service employees who wish to remain anonymous. The military itself denies the report.

“Lavender” is said to have classified 37,000 people as suspected members of Hamas or Islamic Jihad shortly after the Hamas massacre. This is what the Israeli-Palestinian media report says. They also report an error rate of around 10 percent, which is several thousand possible errors in judgment. In addition, there is the priced-in collateral damage: the numerous civilian victims who lose their lives when “Where’s Daddy” successfully locates a person and a bomb hits a residential building.

Where intelligence analysts once managed a few dozen carefully researched targets per week, AI-powered software like Lavender now delivers thousands of targets in a very short time. The danger: In war, artificial intelligence becomes a machine that writes death lists and flies to targets so quickly that checking and control by humans becomes fiction.

The effects of AI weapon systems are not only devastating in Gaza. In Iran, for example, dozens of children and teachers died in the spring of 2026 when a girls' school was attacked by AI technology from the USA because the school was presumably coded as a military object in an outdated database. In the Sudan war, cheap, autonomous drones are used in densely populated areas where it is difficult to distinguish between civilians and combatants. The list goes on.

Today, large language models, i.e. AI models that process human language and generate texts, advise soldiers. Navigation algorithms help approach military targets, and decision-making algorithms suggest attacks. Unmanned kamikaze drones and other so-called killer robots serve as weapons. The latter are called “LAWS” in military jargon, lethal autonomous weapon systems. They are highly complex systems in which AI is used to varying degrees and which can also kill independently.

AI weapon systems have far-reaching consequences, but comprehensive rules for their use are still missing. While Austria is considered one of the pioneers in banning high-risk, high-tech weapons, Germany remains conspicuously silent when it comes to regulating automated weapon systems.

George-Wilhelm Gallhofer is head of the disarmament department at the Austrian Foreign Ministry. “We launched the first UN General Assembly resolution in 2023 to raise awareness and advance the regulation of LAWS,” says Gallhofer. Nevertheless, there is currently neither an international understanding of principles nor a resolution that has been agreed upon internationally. However, there are ideas and ways to achieve binding bans on autonomous weapon systems that are considered particularly risky. What do these look like? And what's wrong so far?

The ban debate is not entirely new. From nuclear weapons to landmines: the global community has already found regulations for some high-risk weapons. However, this often only followed a certain amount of pressure, for example when the shock of the Cuban Missile Crisis gave rise to urgent reasons for restricting nuclear weapons.

But today things are different: landmines can be counted and stocks can be destroyed. AI software in military equipment, on the other hand, is invisible and can be copied with little effort. And: AI autonomy is a spectrum - the question of when a system is autonomous alone is already dividing negotiations. Regulation that takes measures based on the current situation always lags behind the real situation. The pace of development of autonomous systems is literally increasing. Experts speak of a spiral of armament: states are continually expanding their weapons stocks. The credo, especially of the leading industrialized countries, seems to be: Innovation first - and please with us!

You are reading a text from our future department. If you're interested in more positive perspectives, subscribe to TEAM FUTURE, the constructive newsletter on climate, knowledge, utopias. Every Thursday you will receive an email from us with strong thoughts for you and the planet.

🐾 Sign up now for the TEAM FUTURE newsletter – free of charge

More and more money is flowing into the defense sector. The Bundeswehr will have almost 108 billion euros in 2026, and in 2027 it will be almost 140 billion euros. Not only established companies such as Rheinmetall or Hensoldt benefit from this, but also start-ups that develop various autonomous weapon systems.

Until before the Russian war of aggression, armaments were considered rather unpopular among investors. But now young companies are completing large rounds of financing. They develop weapons systems before the federal government has even ordered them, partly financed by right-wing libertarian forces like US tech billionaire Peter Thiel. Do we really want to leave it up to the moral compass of tech enthusiasts and entrepreneurs to decide how deadly AI wars could become?

According to the Geneva Conventions, the final decision whether to kill people or destroy objects when using weapons must rest with people, says Jutta Weber, professor of media sociology at the Institute for Media Studies at the University of Paderborn and expert in AI-based warfare.

An example from history shows that humans are indispensable as a corrective. On September 26, 1983, a Soviet early warning system sounded the alarm. Five US nuclear missiles are said to be approaching. The regulation required that the attack be reported immediately in order to launch a devastating counterattack. But the officer on duty hesitated. Only a handful of rockets for an initial attack seemed unusual to him. So, contrary to the system's recommendations, he did not report the alleged attack. In this way he possibly prevented a nuclear war.

The “human-in-the-loop” approach is now considered the gold standard for AI weapon systems. This means that a human being is involved in the process. His judgment is in demand and he supervises processes. It is at least a rule that many countries have been able to agree on so far. But the protective effect is limited: even if a human is involved, partial automated decisions have already been made that they can hardly understand. And many systems are already designed to function completely independently in the future.

The Geneva Conventions, the heart of international law, cannot be implemented properly, warns Jutta Weber in an interview with the taz: “If decisions are made at machine speed, it cannot work.”

How does the global community discuss AI weapon systems? It is worth looking at the top of the United Nations. Since the beginning of his term in office in 2018, UN Secretary-General António Guterres has been faced with a small but remarkably stubborn crowd: a minority of the 193 member states are blocking strict rules or even a ban on certain AI weapons. The main blockers include the USA, Russia, India and Israel.

Just a few weeks ago, Guterres appealed to the United Nations: “We are dangerously close to crossing a moral red line: the autonomous targeting of humans by machines.” It is not his first warning to political decision-makers. But it is more urgent than ever before; at a certain point the technological spread and geopolitical dynamics can no longer be regulated. This is also emphasized by Gallhofer, the head of the disarmament department in Austria.

The majority of states understand this; the nations represented would be quite capable of enforcing internationally binding regulations for the use of AI. The European Parliament is calling for a complete ban on certain weapons. The human rights organization Amnesty International is of the opinion that “the use of autonomous weapon systems to directly use force against people can never be compatible with international human rights standards and humanitarian law.” The Vatican wants a temporary halt to the use of AI in war until there is a legally binding global AI weapons system regulation.

The countries that are at odds over the UN Convention on Weapons (CCW) at the UN office in Geneva have one thing in common, says Thomas Küchenmeister from the “Stop Killer Robots” campaign. “The countries that are blocking are the technology leaders and users of AI-powered, autonomous systems.” Küchenmeister is the coordinator of the German branch of the international campaign for a global, internationally binding ban on fully autonomous weapon systems. He says: “After black powder and nuclear weapons, autonomous weapons represent the third revolution in warfare.” However, the countries that were slowing down insisted on non-binding codes of conduct. AI regulation is being discussed, but with as little impact as possible. “Agreements have been reached at the lowest, most non-binding level,” says Küchenmeister about the United Nations.

By the way: The federal government has been conspicuously reserved in the past when it comes to AI restrictions. A coalition promise to address the issue has been in place since 2013. The “Stop Killer Robots” campaign criticizes the fact that it is repeatedly broken. The CDU/CSU, FDP and AfD reject a complete ban on AI-supported weapons. As one of the last NATO members, Germany was without an AI strategy for its military until a few months ago. The AI ​​concept the Bundeswehr has just presented does not go beyond general ethical principles, says Küchenmeister. There were no clear limits on the autonomous use of force, verifiable security standards, transparent responsibility structures and guidelines for effective independent control.

In the debate about bans and regulation, the West could learn from countries in the so-called Global South, says Ishmael Bhila. He is a peace researcher at the University of Paderborn. As early as 2025, 49 of 54 African states supported the latest United Nations resolution on completely self-sufficient AI weapons. At the same time, own approaches have been formulated for years, as some countries in the Global South are importers rather than developers of autonomous weapon systems and/or are particularly exposed to the use of these weapons. Since AI weapons are used in wars here, the resulting damage is immediately noticeable. Sierra Leone submitted a joint declaration for a binding instrument on behalf of eleven states in 2021. The Non-Aligned Movement of 120 countries that did not want to join the Western camp or the Eastern Bloc during the Cold War also called for a binding instrument in 2022.

There are viable options for an international agreement on autonomous weapon systems and the ban on LAWS: One of these is an agreement within the framework of the UN Weapons Convention CCW in Geneva. There has already been impetus for this: “A group of 16 countries from the Global South suggested that we introduce a sixth protocol on autonomous weapon systems in the CCW,” says peace researcher Bhila. So far there have been five protocols, for example on landmines, biological and laser weapons.

Protocol VI would regulate how autonomous weapon systems are defined and regulated - and, more importantly, when they are banned. This would have the advantage that almost all relevant military powers would be at the table from the start: the USA, Russia and China. The agreement would be internationally robust. However, there is a major hurdle because the process works based on a consensus principle. If only one nation votes with a veto, the binding regulations are blocked. “That means that countries that are currently attacking others would also have to agree,” says Bhila. The prospect of reaching such an agreement is slim, to say the least. After all, people have been trying for over a decade.

There has now been a small success at the UN level. On September 5, 2026, 128 nations agreed on a preliminary common definition of LAWS for the first time. Germany also voted for a common basis for negotiations on an agreement that is binding under international law. This definition was urgently needed so that the next negotiation could begin. From November 16th to 20th, delegates in Geneva will attempt for the seventh time what has failed six times before: establishing international commitments.

This text first appeared in wochentaz, our weekly newspaper from the left!

Every week, wochentaz is about the world as it is – and as it could be. A left-wing weekly newspaper with a voice, attitude and the special taz view of the world. New every Saturday at the kiosk and of course by subscription.

If consensus is not possible in Geneva, there is a second option for international rules. The prime example of this is the Ottawa Convention. In the fall of 1996, the international community was stuck. In the Geneva negotiations on a mine ban, the consensus principle blocked any progress. But something had been building up outside the conference halls. The International Campaign to Ban Landmines (ICBL), founded in 1992, played a key role, an alliance of aid organizations that brought together victims, deminers and doctors and had prominent support. Images of Princess Diana walking through a minefield in Angola went viral.

The key moment came at a conference in Ottawa: Canada's Foreign Minister Lloyd Axworthy called on the states not to wait any longer for Geneva and to sign a treaty in Ottawa within a year. 14 months later, in December 1997, the time had come. The Ottawa Convention is considered a milestone in humanitarian disarmament. A coalition of the willing became an international treaty that outlaws landmines. Over 160 countries have now ratified the agreement, and several million stored landmines have been rendered harmless.

What worked then could also work today. States come together independently outside of the consensus bodies in order to escape the blocked UN forum. The advantage would be that an agreement could be implemented much more quickly and would increase the pressure on non-signatories. However, there are also hurdles: the route via Ottawa worked because implementing the ban cost comparatively little. In addition, the goal of banning anti-personnel mines was much clearer than it is today. In contrast, LAWS, i.e. high-risk AI weapon systems, are defined, among other things, by their degree of autonomy. Agreeing on what exactly should be banned is more difficult to grasp and the enforcement of a ban is also more difficult to control later. What's more, the major powers would most likely stay away from the negotiations. This means: The effect would initially be much more symbolic and less binding for the developers of so-called AI killer robots.

Whether the end result is an additional protocol for dealing with AI weapon systems or a treaty that some states conclude bypassing Geneva is secondary. What matters is what it says. An agreement worthy of the name must outlaw systems that cannot be used in accordance with international law and, for all others, define what actually constitutes “significant human control”.

If the negotiations are moved away from the consensus-based UN Weapons Convention to the UN General Assembly in New York City, a third path opens up for the global community; here a resolution can be passed with a simple or two-thirds majority. That would create political pressure and could be implemented much more quickly. However, such a resolution would not be legally binding and a further step would be needed for a real treaty with ratification requirements. The impetus would have to come from pioneers among the UN members, such as Austria, which has already distinguished itself in the past by initiating negotiations.

The term Oppenheimer moment was coined when Oppenheimer, the inventor of the nuclear weapon, realized its deadly effects. Since then, it has been used for new, particularly powerful technologies that can fundamentally change the world and whose control is slipping away from people. When it comes to AI weapons, it will soon become clear what the current global community will do with its Oppenheimer moment.