Artificial intelligence is becoming increasingly embedded in military operations around the world, with experts warning that the technology could accelerate decisions in war, influence human commanders and, in extreme scenarios, contribute to nuclear escalation.
“AI is being integrated into militaries globally at an increasingly rapid pace as many governments see AI as decisive for their military competitiveness,” Georgia Cole, a research fellow in Chatham House’s International Security Programme, told The Independent.
Cole said AI was being deployed across the military spectrum, including battlefield awareness, training, surveillance and reconnaissance, target identification and unmanned military vehicles. The source cites Russia’s deployment of a fully autonomous drone that killed three Ukrainian civilians as one example of how autonomous systems are being used in the war in Ukraine.
The expansion of military applications comes as governments and technology companies grapple with wider concerns about the risks posed by increasingly capable AI systems. Earlier this month, an Anthropic researcher warned that there was a 10 per cent chance artificial intelligence could destroy humanity. US President Donald Trump has separately declared: “Whoever wins AI, wins!” He has said that he and the US are the only guardrails needed for the technology, while referring to artificial intelligence as “super intelligence”.
The source also describes instances in which AI systems have been used or tested for activities connected to conflict and security. In 2024, reports alleged that Israel’s military had used an AI-based programme known as Lavender in Gaza following the 7 October attacks by Hamas. The Israel Defence Forces denied using AI to generate confirmed military targets, while confirming that Lavender had been used to “cross-reference intelligence sources” and produce updated information on military operatives of terrorist organisations. The IDF said its information systems were tools for analysts involved in identifying targets.
More recently, according to a threat intelligence report released by Anthropic, AI users in Houthi-held Yemen attempted to use Claude to obtain instructions for building ballistic missiles, while Russian hackers tried to use the technology to spy on groups in Ukraine. Anthropic said biological misuse was among the most serious risks posed by frontier AI models, but did not identify the countries or research institutions involved because it was uncertain about the intentions behind the questions.
The most severe concerns centre on nuclear weapons. Studies cited in the source, including research by King’s College London in February and Stanford University in 2024, found during simulated military crises that AI systems were more likely than humans to escalate towards nuclear threats. The King’s study found that AI chose nuclear signalling in 95 per cent of simulated crises.
Cole said a system could potentially misinterpret information indicating that a nuclear attack was under way and contribute to a retaliatory decision based on an attack that had not occurred. She also warned that AI could reduce transparency and shorten the time available to decision-makers. Systems could also be targeted by state or non-state actors seeking either to disable nuclear capabilities or trigger their use.
At the same time, some of the most dramatic predictions about AI-driven catastrophe are disputed. Yasir Atalan, deputy director and data fellow at the Futures Lab at the Centre for Strategic and International Studies, described the 10 per cent extinction-risk figure cited by the Anthropic researcher as a “red herring” that was not based on “credible analysis”. He said extreme scenarios involving rogue “killer robots” remained doomsday assessments without mechanisms demonstrating how they could occur.
Atalan’s research points to a more immediate concern: AI may alter human behaviour during crises. In a CSIS study, military officers were shown recommendations from AI models, with some receiving escalatory advice and others non-escalatory options. “What we realised is that it actually changed how people make a decision,” Atalan said. “If they get a recommendation from an escalatory model they actually make more escalatory choices.”
AI could also be used by rogue actors for disinformation and misinformation, Atalan said, citing AI-generated Lego videos as part of Iran’s communications and psychological strategy. Cole warned separately that AI agents could potentially hack critical infrastructure at a scale capable of causing mass casualties or assist in developing weapons for asymmetric warfare.
Both experts said the central issue was the safeguards governing military AI rather than the technology itself. Atalan warned against unchecked models being integrated into command-and-control systems, while Cole said there was a need for greater transparency and scrutiny of where AI improves military decision-making and where it causes harm. Discussions are continuing at the United Nations over a treaty on lethal autonomous weapons, but the source says agreement is unlikely soon and that several major powers oppose legally binding limitations.
