We can't let them hide behind their models From Gaza to Iran, the pattern is the same: precision weapons, chosen blindness, and dead children. There is an Israeli military strategy called the "fog procedure". First used during the second intifada, it's an unofficial rule that requires soldiers guarding military posts in conditions of low visibility to shoot bursts of gunfire into the darkness, on the theory that an invisible threat might be lurking. It's violence licensed by blindness. Shoot into the darkness and call it deterrence. With the dawn of AI warfare, that same logic of chosen blindness has been refined, systematized, and handed off to a machine.

As military organisations consider integrating large language models (LLMs) into command and control (C2) systems for planning and decision support, understanding their behavioural tendencies is critical. This study develops a benchmarking framework for evaluating aspects of legal and moral risk in targeting behaviour by comparing LLMs acting as agents in multi-turn simulated conflict. We introduce four metrics grounded in International Humanitarian Law (IHL) and military doctrine: Civilian Target Rate (CTR) and Dual-use Target Rate (DTR) assess compliance with legal targeting principles, while Mean and Max Simulated Non-combatant Casualty Value (SNCV) quantify tolerance for civilian harm. We evaluate three frontier models, GPT-4o, Gemini-2.5, and LLaMA-3.1, through 90 multi-agent, multi-turn crisis simulations across three geographic regions. Our findings reveal that off-the-shelf LLMs exhibit concerning and unpredictable targeting behaviour in simulated conflict environments. All models violated the IHL principle of distinction by targeting civilian objects, with breach rates ranging from 16.7% to 66.7%. Harm tolerance escalated through crisis simulations with MeanSNCV increasing from 16.5 in early turns to 27.7 in late turns. Significant inter-model variation emerged: LLaMA-3.1 selected an average of 3.47 civilian strikes per simulation with MeanSNCV of 28.4, while Gemini-2.5 selected 0.90 civilian strikes with MeanSNCV of 17.6. These differences indicate that model selection for deployment constitutes a choice about acceptable legal and moral risk profiles in military operations. This work seeks to provide a proof-of-concept of potential behavioural risks that could emerge from the use of LLMs in Decision Support Systems (AI DSS) as well as a reproducible benchmarking framework with interpretable metrics for standardising pre-deployment testing.

Large Language Models (LLMs) are widely used across sectors, yet their alignment with International Humanitarian Law (IHL) is not well understood. This study evaluates eight leading LLMs on their ability to refuse prompts that explicitly violate these legal frameworks, focusing also on helpfulness - how clearly and constructively refusals are communicated. While most models rejected unlawful requests, the clarity and consistency of their responses varied. By revealing the model's rationale and referencing relevant legal or safety principles, explanatory refusals clarify the system's boundaries, reduce ambiguity, and help prevent misuse. A standardised system-level safety prompt significantly improved the quality of the explanations expressed within refusals in most models, highlighting the effectiveness of lightweight interventions. However, more complex prompts involving technical language or requests for code revealed ongoing vulnerabilities. These findings contribute to the development of safer, more transparent AI systems and propose a benchmark to evaluate the compliance of LLM with IHL.

At least one person was injured and several homes damaged in a Russian drone attack on Ukraine's Kyiv region, Governor Ruslan Kravchenko said. Ukraine's Air Force said it shot down 53 of the 56 Russian drones that targeted the country's central, northern and southern regions. Air defence units destroyed nearly 20 drones that were heading towards Kyiv itself, the military said. Ukrainian President Volodymyr Zelenskyy, speaking in his nightly video address, said there had been 100 battles over the past 24 hours on the eastern front with the heaviest fighting in the Pokrovsk and Kurakhove sectors. Russia ordered the evacuation of settlements close to the Ukrainian border in the Kursk region and said it had retaken two villages – Uspenovka and Borki – Ukraine captured last month in a surprise cross-border incursion.

It has now been close to 10 months that Israel has been waging a genocidal war in Gaza. Its army has violated nearly every facet of international humanitarian law in its relentless assault on an unimaginably vulnerable population. Israel has denied the Gaza concentration camp the bare necessities of life -- food, water, medicine, sanitation, electricity and fuel. And its targeting of civilian infrastructure has made the majority of Gaza residents homeless. No Israeli military goal requires the total destruction of Gaza.

Fears the Pentagon has been'building killer robots in the basement' may have led to stricter AI rules that mandated all systems must be approved before deployment. The Department of Defense (DoD) recently updated its AI rules among'a lot of confusion about' how it plans to use self-decision-making machines on the battlefield, according to the deputy assistant defense secretary. Michael Horowitz explained at an event this month that the'directive does not prohibit the development of any systems,' but will'make clear what is and isn't allowed' and uphold a'commitment to responsible behavior,' as it develops lethal autonomous systems. While the Pentagon believes the changes should ease the public's minds, some have said they are not'convinced' by the efforts. Fears the Pentagon has been'building killer robots in the basement' may have led to stricter AI rules that mandated all systems must be approved before deployment. News of the update to the Pentagon's 2012 'Autonomy in Weapon Systems,' has sparked a debate online with many people saying'If the Pentagon says they're not doing it, they're doing it.'

United Nations delegates are currently meeting to debate possible regulations controlling autonomous killer robots -- but Russia is having none of it. The Russian delegate, representing a country that has already developed and deployed military robots in real-world conflicts, remained steadfast that the global community doesn't need any new rules or regulations to govern the use of killer robots, The Telegraph reports. That pits Russia against much of the rest of the international community, who are calling for rules to keep humans in charge of the decision to open fire, highlighting on the main anxieties and ethical conundrums surrounding autonomous weaponry. The argument from Russia is that the AI algorithms driving these killer robots are already advanced enough to differentiate friend from foe from civilian, and that therefore there's no need to burden the autonomous death machines with unnecessary regulations. "The high level of autonomy of these weapons allows [them] to operate within a dynamic conflict situation and in various environments while maintaining an appropriate level of selectivity and precision," the delegate said, according to The Telegraph.

Artificial intelligence (AI) is bound to enable innovation in the decades to come. On the one hand, AI technologies may be used to improve societal well-being and help fight human rights abuses. On the other hand, AI presents a variety of challenges that can profoundly affect the respect for and protection of human rights. Therefore, it is important to place international human rights law (IHRL) at the centre of discussions about AI governance. Our New Research Brief Human Rights and the Governance of Artificial Intelligence discusses the opportunities and risks that AI represents for human rights, recalls that IHRL should occupy a central place in the governance of AI and outlines two additional avenues to regulation: public procurement and standardization.

Artificial intelligence developer DeepMind has just announced its latest milestone: a bot called AlphaStar that plays the popular real-time strategy game StarCraft II at Grandmaster level. This isn't the first time a bot has outplayed humans in a strategy war game. In 1981, a program called Eurisko, developed by artificial intelligence (AI) pioneer Doug Lenat, won the US championship of Traveller, a highly complex strategy war game in which players design a fleet of 100 ships. Eurisko was consequently made an honorary Admiral in the Traveller navy. The following year, the tournament rules were overhauled in an attempt to thwart computers.

Members of Congress, the U.S. military, and prominent technologists have raised the alarm that the U.S. is at risk of losing an Artificial Intelligence (AI) arms race. China already has leveraged strategic investment and planning, access to massive data, and suspect business practices to surpass the U.S. in some aspects of AI implementation. There are worries that this competition could extend to the military sphere with serious consequences for U.S. national security. During the prior Cold War arms race era, U.S. policymakers and the military expressed consternation about a so-called "missile gap" with the USSR that potentially gave the Soviets military superiority. Echoes of gap anxiety continue today.