Race Against the Machine

Artificial intelligence has moved far too quickly from laboratory curiosity to military instrument for the world to keep pretending regulation can catch up later. Pakistan's warning at the UN Security Council on Thursday that AI must not become 'another facet of the ongoing arms race' deserves attention precisely because the race is no longer hypothetical. That should ring alarm bells. Deputy Prime Minister Ishaq Dar told the France-chaired meeting that human judgement must remain central to AI and called for guardrails, restraint and confidence-building measures. The danger is already taking shape. AI is being woven into surveillance, targeting, drones, cyber operations and battlefield planning. The frightening question is no longer whether a machine will one day be allowed to kill. It is how much of the chain leading to that decision can be handed to machines before human control becomes a formality. Speed is the real menace. AI can compress the interval between detection and response to seconds. SIPRI has warned that systems outside nuclear command structures can shorten decision-making timelines and increase the risk of miscalculation during a crisis. Between nuclear-armed states, that is not academic. An algorithm that produces a wrong assessment faster is still wrong; humans simply have less time to discover the error. An Iran-linked actor used Anthropic's Claude to gather intelligence and develop targeting recommendations against US naval forces, the company revealed this month. More chillingly, media sources reported last week that an unreleased Pentagon review found overreliance on Palantir's Maven AI system among the failures behind a US strike on a school in Iran that killed at least 123 children. The threat is wider than the battlefield. AI-generated audio, video and images are making deception cheaper and more convincing. Critical infrastructure, financial systems and public opinion can all be attacked without a missile crossing a border. A fabricated order attributed to a military commander during a tense confrontation could do damage before governments establish whether it is genuine. What makes the moment alarming is the contradiction at its centre. Governments speak of safeguards while states and companies compete to build more capable systems. At Thursday's meeting, some frontier-AI developers urged caution. Yet competitive pressure rewards whoever moves first, not whoever stops to ask whether a capability should exist at all. Pakistan is also right that regulation cannot become technological gatekeeping by powerful states or corporations. Developing countries need access to peaceful AI. Access and restraint are not opposites. Civilian innovation can be widened while hard limits are placed around autonomous targeting, nuclear decision-making and attacks on civilian infrastructure. The nuclear age built safeguards after humanity had stared into the abyss. There is no reason to wait for another catastrophe. *