AI enters the war room as diplomacy struggles

Artificial intelligence is rapidly becoming a force multiplier on the battlefield, helping militaries identify targets, guide weapons and compress decisions from minutes to seconds. But while the technology races ahead, diplomacy remains trapped in debates over human control, accountability and rules that may no longer match the realities of modern warfare

Artificial intelligence has moved from the margins of defence planning to the heart of the war room, and the diplomacy meant to govern it is being left far behind. Machines now sift oceans of surveillance data and hand commanders ranked target lists in seconds. The people who set the rules are still arguing over definitions that date back to the last decade.

Washington, Beijing and Moscow treat AI as a strategic asset of the same order as air power or the nuclear arsenal. President Putin said years ago that whoever leads in AI will rule the world. President Xi has called for self-reliance and mastery across every core AI technology. President Trump has committed the United States to doing whatever it takes to lead the field. Those words have turned into budgets. The US Army trains officers on the Maven Smart System, and the Pentagon has created a Defense Autonomous Warfare Group that is absorbing tens of billions of dollars. China and Russia are pouring rising sums into swarming drones and autonomous strike systems, and the two defence ministries have held talks on how AI might aid a large conventional conflict.

The sharpest question is what happens when AI stops driving weapons and starts advising the humans who fire them. At a Congressional hearing this month, witnesses cautioned that an AI system can influence a human decision leading to a strike while the person retains formal responsibility for the final call, and that recommendations can arrive faster than officers can authenticate them, compressing the time to intervene. Responsibility becomes blurred. A commander who approves a target in seconds, trusting a system he cannot fully audit, is accountable in name while the real selection happened inside a model.

Gaza offers the most documented example of what this looks like in practice. Israeli forces used AI systems called Lavender and the Gospel to generate targets at speed, with Lavender marking around 37,000 people as suspected militants and Gospel flagging buildings and infrastructure. A separate tracking tool named “Where’s Daddy?” signalled when a target entered his home, and in some cases the house was marked for bombing while the family was present. Intelligence officers described treating the machine’s output as though it were a human decision, and reported that the army accepted the deaths of 15 to 20 civilians for a junior operative and more than 100 for a senior commander. Legal scholars have identified these practices as potentially violating the principles of distinction, proportionality and precaution.

Ukraine has become the world’s largest live proving ground for the same technology. AI-enabled combat systems there are being tested at a scale seen nowhere else, and Russia has fielded AI-powered strike drones that can evade air defences, identify targets and operate offline. What one war validates spreads quickly to every arsenal watching.

Diplomacy has produced movement without much grip. At their summit in May, the American and Chinese leaders agreed to improve policy exchanges on AI, and a bilateral channel on AI governance was announced, after which Chinese officials urged the world to close the regulatory vacuum around military AI and backed rules that could eventually become binding. Yet both powers sent substantially smaller delegations to the recent Responsible AI in the Military Domain summit in Spain than they had to the 2024 meeting in South Korea. The one understanding they had quietly held, keeping AI away from the decision to use nuclear weapons, looks to be eroding rather than consolidating.

At the United Nations, the picture is slower still. The Group of Governmental Experts spent much of its three-year mandate debating what human control even means and reached the end of that mandate without a binding text, leaving the question to the Seventh Review Conference of the Convention on Certain Conventional Weapons in November. In December 2025, the General Assembly passed Resolution 80/57, and India voted in favour of it. New Delhi holds to a careful line of its own. It backs the expert process and insists that any use of autonomy obey the laws of armed conflict with human accountability intact, while declining to support a legally binding ban. On that stance, it sits alongside the United States, Russia and Israel among the main opponents of treaty negotiations, wary of rules that could freeze its own options.

This is where AI has become a weapons danger of a new kind. The same models that draft text and design molecules can speed the engineering of novel arms and lower the skill needed to build them. Capabilities once held by great powers can pass to smaller states and to groups that answer to no state. Systems that can select and engage targets on their own already exist, are being deployed in combat, and are being procured at unprecedented scale. More than 160 countries have voted for restraint, while the strongest military powers resist anything that might constrain their programmes.

The problem forces a rethink of what a weapon is. For centuries, a weapon waited for a human to aim and fire. A system that picks its own targets breaks that assumption and asks whether a machine should ever be allowed to make the choice that ends a life. Defence ministries insist a person will always stay in the loop. That loop keeps tightening as the systems grow faster, and a human given three seconds to say yes is a supervisor in title alone. For now, the race is ahead of the rulebook by a clear margin. Whether governments can close that gap before an AI-driven error, in a crowded sky or a nuclear command chain, forces a decision no one intended is the test the coming summits and the November review will have to meet. The technology will not wait for them to agree.

At the United Nations, the picture is slower still. The Group of Governmental Experts spent much of its three-year mandate debating what human control even means and reached the end of that mandate without a binding text, leaving the question to the Seventh Review Conference of the Convention on Certain Conventional Weapons in November

The writer is a former Vice-Chancellor of Jamia Hamdard, New Delhi; Views presented are personal.