The REAIM conference attempted to advance rules for the responsible use of AI in weapons systems, but less than half of the countries signed the non-binding declaration. The US and China abstained, and concerns about losing military advantage continue to block global regulation.
The international effort to regulate the use of artificial intelligence in weapons systems has run into a major obstacle: states fear that any restrictions they impose will give an advantage to enemies who do not behave with the same degree of responsibility. An article published in The Conversation by Dr. Mark Tsagas, Senior Lecturer in Law, Cybercrime and the Ethics of Artificial Intelligence at the University of East London, describes how this fear blocks the transition from statements of principle to actual rules of operation.
In February 2026, the third conference in the REAIM series of conferences on responsible artificial intelligence in the military was held in La Coruña, Spain. The aim of the meeting was to translate previously formulated principles into practical steps. It was attended by government representatives, military personnel, industry representatives and researchers from policy institutes. The conference presented a non-binding commitment, which called for maintaining human oversight of military AI systems, for carrying out rigorous risk assessments and testing, and for transparency in decision-making processes when using artificial intelligence in conflicts.
On the surface, these are basic principles that are hard to resist. But in reality, less than half of the countries that participated in the conference signed the pledge. The US and China, two of the most important players in the field, chose not to sign. According to the article, their abstention reinforced the hesitation of other countries and illustrated how difficult it is to reach agreement when technology is seen as part of the struggle for military advantage.
The prisoner's dilemma of the states
The Dutch Defense Minister, Ruben Berkelmans, described the situation as a “prisoner’s dilemma.” It’s a concept from game theory in which two parties may gain from cooperation, but each fears that the other will break the rules and exploit their trust. In the military realm, the implication is clear: a country might agree to responsible restrictions on the use of AI, but fear that a rival country will develop more advanced autonomous systems and use them without the same restrictions.
The problem is particularly acute when it comes to autonomous weapons systems. Such systems, sometimes called “lethal robots,” are of concern because they may make operational decisions without sufficient human intervention. While there is broad consensus against the irresponsible use of such systems, countries have yet to agree on even a uniform definition of them. Without an agreed definition, it is difficult to begin a serious discussion about regulation.
The fear of adopting a binding definition is not just legal. Countries fear that agreeing to a definition that is too broad will limit their armies' freedom of action on the battlefield. That is why attempts by academic experts to propose flexible formulations have so far failed to bridge the gaps.
The battlefields became testing grounds.
The article notes that existing laws of war already prohibit, at least in principle, irresponsible and unethical use of military technologies. However, the question is how rules formulated for a more humane world will operate when intelligence decisions, target identification, or the operation of weapons systems are made with the assistance of artificial intelligence systems.
The war in Ukraine, the war in Gaza, and the escalation in Iran are presented in the article as examples of situations in which armies use new technologies as a kind of testing ground. The article mentions, among other things, systems for collecting intelligence and assisting in identifying targets, as well as advanced uses of AI models for analyzing information. The emphasis is not only on autonomous weapons, but also on the intelligence and planning layer: systems that help commanders make decisions faster, sometimes in an environment of partial information and operational pressure.
At the same time, drone warfare is evolving rapidly. AI-assisted drones, semi-autonomous drones, and systems with more autonomous capabilities are already changing the battlefield. According to the article, the pace of development of the technology is much faster than the pace of formulating the rules that are supposed to regulate it.
Human supervision that is not always significant
One of the main arguments for the cautious use of military AI is to keep a “human in the loop.” That is, to require that critical decisions remain under human supervision. But even this solution is not without its dangers. When operators become accustomed to and trust AI systems too much, they can become passive approvers of the system’s recommendations. This phenomenon is known as automation bias.
In such a situation, human oversight may lose its meaning. Instead of critically reviewing the system's decisions, the human may press approval almost automatically. As AI systems become more deeply integrated into intelligence, planning, and attack processes, this risk becomes more serious.
The article also highlights well-known problems with AI systems: biases, misinformation, disinformation, and the erosion of human judgment due to overreliance on automated systems. When it comes to civilian systems, such malfunctions can cause harm. When it comes to the battlefield, they can lead to unintended escalation and harm to civilians.
The main conclusion is that the international community knows how to identify the risks, but has difficulty turning this recognition into binding rules. As long as countries fear that committing to responsibility will harm them against an adversary who does not abide by the same rules, international regulation of military AI will continue to progress much more slowly than the technology itself.
Full amp
title:
subtitle:
Tags: Artificial Intelligence, Military AI Autonomous Weapons, Weapon Systems, Law of War, Drones, Lethal Robots, Ethics of Artificial Intelligence, International Security, REAIM, Automation Bias
Key phrase: Military artificial intelligence
Synonyms:Military AI, Artificial Intelligence in Weapon Systems, Autonomous Weapons, Autonomous Weapon Systems, Lethal Robots, Artificial Intelligence in War, Military Use of Artificial Intelligence, Regulation of Military AI
SLUG: military-ai-weapons-regulation-prisoners-dilemma
title SEO:
Military Artificial Intelligence: Why the World Is Struggling to Regulate It
Meta description:
Countries fear that limiting military AI will help their enemies. The REAIM conference exposed the difficulty of regulating autonomous weapons and AI on the battlefield.
Keywords:
Military AI, Autonomous Weapons, Military AI Regulation, Lethal Robots, Laws of War, Automation Bias, Military Drones, REAIM
Internal link suggestions:
Artificial Intelligence and Security, Anchor: Uses of AI in Security Systems
Drones and the Modern Battlefield, Anchor: Drone Warfare
Ethics of Artificial Intelligence, Anchor: Ethical Risks of AI
FAQ short:
What is military artificial intelligence??
Military artificial intelligence is the use of AI systems for intelligence, planning, information analysis, target identification, or the operation of weapons systems.
What is the main problem in regulating AI military?
The problem is that countries fear that the restrictions they impose will be exploited by adversaries who do not abide by the same rules.
What is automation bias??
Automation bias is a situation where humans place too much trust in the recommendations of a computer system, to the point that they approve them without sufficient critical scrutiny.
