The use of artificial intelligence (AI) has increased significantly over the past few years across various fields, and war is no exception. Autonomous weapon systems (AWS), also called ‘killer robots’, combine AI with military equipment and can be used on land, in air, in water and even in outer space. They can be lethal, such as drones that attack enemy targets without human intervention, or non-lethal, such as surveillance equipment. Their rise has raised questions about legality and ethics, in particular whether they comply with International Humanitarian Law (IHL) and customary law.
Militaries are heavily investing in AWS because these machines can identify and attack a target on their own. This reduces the need for human combatants, saving lives and costs, and machines are arguably faster, more flexible and more efficient. The US Deputy Secretary of Defense, Bob Work, even envisions “NATO’s military superiority” through their development.
The advantages are real. AI can support decision making, whether in planning military operations or carrying them out. Drones can capture images and transmit them for analysis, making it easier to gather geospatial intelligence on a target area and its terrain. Autonomous or partially autonomous assistants can dismantle bombs and mines, evacuate the injured, explore areas inaccessible to humans and deliver supplies such as medicine and food. Shorter decision times can also lead to more efficient action in high-pressure situations.
The dark side is total autonomy. A war on autopilot would mean great destruction and would be difficult to stop once started. AI can also locate high value military assets, which becomes dangerous if it is used to find nuclear weapons, since an attack on them could cause devastation for many miles beyond the target. Still, fully autonomous weapons may not be the whole problem. Even with a human taking the final decision, there is room for abuse, and human oversight does not guarantee lawful conduct. The issue is not the technology itself but the way it is, or could be, used. This is why legal principles must be formed, compliance mandated and punitive measures enforced.
The debate is deeply divided. NGOs such as Human Rights Watch and Amnesty International call for a ban, citing lack of accountability and mistakes made by machines. Governments and militaries, on the other hand, see AWS as part of the ‘revolution in military affairs’ (RMA), emphasising precision, shorter decision time and fewer lives lost on the frontlines.
At the heart of the legal problem is accountability. AWS can target and kill without human intervention, so responsibility can no longer be assigned neatly to an officer, a state or someone up the chain of command. Gunther Teubner’s idea of ‘network share liability’ offers one answer: instead of blaming a single person or entity, responsibility is shared across the network of people involved, from engineers and programmers to those who deployed the machine. Since machines lack legal personhood and intent, blaming the machine is not an option. Scholars argue that AWS are lawful only with meaningful human oversight.
Existing law already applies. The Geneva Conventions, their Additional Protocols and the Hague Conventions limit the methods and means of warfare. Article 36 of Additional Protocol I requires any new weapon to be legally reviewed, and Article 35 forbids unnecessary suffering. Article 48 sets out the principle of distinction between combatants and civilians, Article 51 covers proportionality, and Article 57 requires precautions in attack. None of these mention AI, but they apply to all weapons. A 2021 SIPRI study adds that AWS can only be lawful if their effects are predictable, controllable and traceable so that humans can be held accountable. If even one condition fails, their use violates IHL.
The ICJ’s 1996 advisory opinion on nuclear weapons offers a useful analogy. The Court found that the absence of an explicit prohibition does not make a weapon lawful, and that weapons unable to distinguish between targets and civilians may be unlawful. AWS cannot yet make this distinction with a high degree of accuracy, and Article 57 requires situational awareness and human judgement, which AI does not guarantee.
The Martens Clause fills the remaining gap. Introduced in the 1899 Hague Convention and later adopted into the Geneva Conventions and Additional Protocol I, it provides that where no specific law exists, civilians and combatants remain protected by the principles of humanity and the public conscience. Since no treaty yet regulates AWS, the clause acts as a moral safeguard: deploying them in a way that violates human dignity would breach international law even without an explicit ban.
States differ on regulation. The United States has no AWS-specific statute, and its 2023 Directive 3000.09 is encouraging towards AWS while promising human oversight. It also opposed a treaty on the issue at the UN General Assembly. The United Kingdom likewise has no specific laws but insists on compliance with IHL. Pakistan has advocated for binding rules, arguing that IHL does not address every challenge AWS pose.
Experts say that deploying AWS may not be illegal in itself, but the way they operate may violate IHL. Extending old laws to new technology may not be enough. A dedicated framework with strict enforcement is needed, one that balances a state’s right to defend itself with the principles of humanity.
