Category: News | Published: 2026-06-09
For most of the past eight decades, nuclear weapons have sat at the top of the strategic risk hierarchy. The logic of deterrence, the careful management of decision timelines, and the extensive protocols built around nuclear command and control all exist because the consequences of getting things wrong are too severe to leave to chance.
Something has shifted. At one of the world's most significant defence and security forums, senior military leaders from across the globe ranked artificial intelligence as a bigger concern than nuclear weapons. Not as a future worry, but as a present and accelerating AI defence risk that is already reshaping how conflicts are fought and how quickly they can escalate.
The Shangri-La Dialogue
The Shangri-La Dialogue is held annually in Singapore and brings together defence ministers, senior military commanders, policymakers, and security experts from across the Indo-Pacific region and beyond. It is one of the most important forums in the world for candid discussions about strategic stability and security risk.
Historically, those discussions have centred on nuclear deterrence, missile defence, arms control agreements, and the balance of power between major states. This year, artificial intelligence dominated the agenda in ways that previous years had not seen. The concerns raised were practical, specific, and came from people who are responsible for managing real military capabilities.
The OODA Loop Problem
To understand why military leaders are worried about AI defence risks in the way they are, it helps to understand a concept called the OODA loop. OODA stands for Observe, Orient, Decide, and Act. It is the basic cycle through which military decision-makers process information and respond to developing situations. The faster you can complete that loop, the greater your operational advantage.
AI is compressing that cycle dramatically. Lieutenant General Nauman Zakria, Commander of 1 Corps and the Army Rocket Force Command of the Pakistan Army, explained the concern clearly at the forum. AI is now accelerating the OODA loop to the point where, in his words, a human simply cannot evaluate the situation fast enough.
His warning about what follows from that was stark. When decision-makers cannot keep pace with the speed of the systems they are supposed to be overseeing, he said, people will act irrationally, and the actions will be extreme.
That is not a theoretical scenario. It is a description of what happens when oversight breaks down because the speed of events exceeds human capacity to follow them.
Speed as a Source of AI Defence Risk
The danger here is not that AI systems become malicious. It is something more mundane and in many ways more worrying: speed-driven miscalculation.
For decades, strategic stability has rested on the assumption that there is enough time between an incident occurring and a response being required for human decision-makers to consult advisers, gather information, communicate with allies, question assumptions, and consider consequences. That time buffer is what turned the Cuban Missile Crisis into a negotiation rather than a catastrophe. The slow pace of decision-making in that situation, which at the time felt agonising, was actually what kept it from escalating to catastrophe.
Research published by SIPRI in June 2025 found that AI-driven decision-making systems can reduce response times by up to 80 per cent. The same research found that false alarm rates in AI-assisted systems increase by 200 per cent compared with conventional systems. Put those two findings together and you have a situation where decisions are being made much faster, based on information that is significantly more likely to be wrong.
The US military's Mosaic Warfare concept, which aims to compress the entire process from situational awareness to strike decision into a matter of minutes, illustrates how seriously military organisations are pursuing this kind of speed advantage. The problem is that when the decision cycle operates at that pace, the opportunity to catch and correct errors before they trigger a response disappears.
Already Happening
One of the most important points made at the Shangri-La Dialogue was that these AI defence risks are no longer hypothetical. General Onno Eichelsheim, Chief of Defence of the Netherlands, pointed to active deployments in the conflict in Ukraine, where AI systems have been used to anticipate Russian attack patterns and coordinate drone operations. The United States has also confirmed that it is using AI tools to support military planning and targeting decisions.
Eichelsheim described AI as a huge risk in escalation and was unambiguous that this concern is already well founded. But he also acknowledged the bind that military organisations find themselves in. Stepping back from military AI is not a realistic option when adversaries are not doing the same. As he put it, the technology will be used in the domain, and it already is.
That combination of recognised risk and continued adoption is what makes this a genuinely difficult AI defence problem rather than one that can be resolved simply by choosing not to use the technology.
New Vulnerabilities Military AI Creates
Beyond the speed and escalation risks, AI systems introduce a category of vulnerability that is entirely new to military planning. Conventional weapons can be physically secured. AI systems can be attacked in entirely different ways.
Data poisoning involves corrupting the training data that AI systems learn from, causing them to develop subtly wrong models of situations and threats. Model hijacking involves manipulating an AI system's inputs to cause it to behave in ways its operators did not intend. Both represent AI-specific attack vectors that existing military security frameworks were not designed to address, and both could in principle cause an AI-assisted system to misclassify a situation and generate a response that escalates rather than manages a conflict.
The Autonomous Weapons Market Is Growing Fast
The scale of investment in military AI makes these AI defence risks more pressing with each year. The global market for lethal autonomous weapons systems reached $14.2 billion in 2024 and is projected to reach $33.47 billion by 2032, growing at a compound annual rate of more than 11 per cent.
That pace of growth reflects how seriously military organisations around the world are treating AI-enabled autonomous systems as a strategic priority. It also means that the window for establishing governance frameworks, oversight protocols, and international agreements around AI defence risk is narrowing.
The integration of AI with robotics and unmanned systems is accelerating that process. US startup Foundation Future Industries recently conducted trials of humanoid robots in Ukraine, with logistics applications as the initial focus. The company has indicated it hopes to move to more advanced military deployments within the next 12 to 18 months. Whether or not those specific plans materialise, they illustrate the direction of travel.
The Humanitarian Dimension
While military leaders focused primarily on strategic and operational risks, Mirjana Spoljaric, President of the International Committee of the Red Cross, raised the humanitarian dimension. Her concern was the growing distance that automated systems place between the human decision-maker and the consequences of military action.
When the chain from decision to impact becomes increasingly automated, accountability becomes harder to establish and the inhibitions that human presence in a situation normally creates are weakened. That raises profound questions about how the laws of armed conflict apply in an environment where AI systems are making or substantially influencing decisions that have lethal consequences.
Those are questions that do not have settled answers, and the pace at which the technology is being deployed is outrunning the pace at which the legal and ethical frameworks are being developed.
Why This Matters Beyond the Battlefield
These discussions might seem distant from the concerns of a typical business. But the AI defence risk debate reflects a broader truth about AI that applies across sectors: the faster AI systems operate and the more consequential their outputs become, the more important it is to maintain meaningful human oversight.
In cybersecurity, AI tools are already being used on both sides. Attackers use AI to find vulnerabilities, craft phishing content, and automate intrusions. Defenders use AI to detect threats, analyse anomalies, and respond to incidents faster than human analysts could alone. The escalation dynamic in that environment is not entirely unlike what military leaders are describing: speed and automation create advantages, but they also create new risks and new failure modes.
Building AI into your organisation thoughtfully, with clear governance, human oversight, and an understanding of what the tools are actually doing, is the right approach whether you are thinking about cybersecurity or business operations more broadly. Our Cyber Security page covers the range of protections we help businesses put in place as the threat landscape continues to evolve.
The Question That Needs Answering
The Shangri-La Dialogue did not produce a resolution to the AI defence risk challenge. What it produced was a clear signal that the people most responsible for managing these risks are genuinely concerned, and that those concerns are grounded in what is already happening rather than what might happen in the future.
The question of how to preserve meaningful human oversight as AI systems operate at speeds that exceed human decision-making capacity is one of the most important questions in security today. The fact that it has overtaken nuclear weapons as the primary concern of senior military leaders tells you something important about where we are.