Europe’s relationship with artificial intelligence in warfare lived for years in the subjunctive mood. Parliament resolutions in 2018 and 2021 called for a global ban on lethal autonomous weapons lacking “meaningful human control.” The European Defence Fund wrote that same phrase into its grant conditions. Brussels positioned itself as the conscience of a world racing toward machine-speed killing. None of it had been tested against an actual war. February 28, 2026 ended the comfort. The United States and Israel launched a joint campaign against Iran, killing Supreme Leader Ali Khamenei within hours and striking roughly a thousand targets generated by Palantir’s Maven Smart System, running on Anthropic’s AI model, in the first twenty-four hours alone, according to Responsible Statecraft. By April, the White House cited more than 13,000 targets struck, the Arms Control Association reported. Among the first day’s targets sat the Shajareh Tayyebeh elementary school in Minab, a former military facility long since converted to civilian use. At least 165 people died, most of them children, and nobody running the system can say where the algorithm’s judgment ended and a human being’s began.
European anxiety about AI in combat used to live mostly in the future tense, fixed on hypothetical drone swarms choosing their own targets. Israel’s AI-assisted targeting in Gaza, reported well before the Iran war, registered in European debate but rarely broke through as urgent, as an earlier Euronews investigation into Mossad’s own AI use suggested. Iran changed the calculus through scale and proximity. This was a full state-on-state war across a country the size of France and Germany combined, not an asymmetric campaign confined to 365 square kilometers of Gaza. CENTCOM commander Admiral Brad Cooper confirmed publicly that AI tools compressed targeting decisions once requiring hours into seconds, even as he insisted humans retained final authority, Al Jazeera reported. Researchers including Newcastle University’s Craig Jones have noted no real evidence exists that such acceleration makes war more humane, and considerable evidence it may not, he told Democracy Now. The Minab strike gave that abstract worry a death toll and a date.
A second shock landed closer to home, and cut deeper. NATO quietly acquired its own version of Maven from Palantir in 2025, meaning the architecture under scrutiny in Iran already sat inside European militaries, outside the reach of the EU’s own AI Act, according to the Centre for Future Generations. The regulation deliberately exempted military and national security uses when it was negotiated, deferring instead to international humanitarian law and member-state discretion, a carve-out France pushed hardest for. Europe spent years building the world’s most elaborate civilian AI rulebook while leaving its own war-fighting algorithms essentially unregulated by anything binding.
June 12, 2026 made the dependency concrete. The US Commerce Department ordered Anthropic to cut off access to its most capable models for all non-US persons, severing European governments and militaries from infrastructure many had quietly built operations around, with no consultation and no avenue for objection, CEPR researchers found. Economists there called it a sovereignty story rather than a regulation story, estimating a decade or more and tens of billions of euros before Europe could negotiate as a peer rather than a dependent. Frontier AI access, unlike energy or semiconductors, can apparently be switched off globally in an afternoon by administrative fiat.
Together these two shocks produced a genuine shift in perception. AI in warfare stopped being a problem for the 2030s and became a problem that had already killed a school full of Iranian children and proved Europe’s defense establishment could be cut off from its own targeting infrastructure overnight. Polling shows large majorities across Western Europe opposing the war outright, and unfavorable views of Israel running as high as 63 to 70 percent, concentrated even more heavily among younger Europeans, Responsible Statecraft noted. The public got there faster than its governments did.
Here the gap between rhetoric and conduct becomes difficult to excuse. Nearly every serious international lawyer examining the February 28 strikes agrees they meet the UN Charter’s definition of an unlawful use of force, authorized by no Security Council resolution and responding to no imminent armed attack, the European Council on Foreign Relations has documented. Carl Bildt said so outright, as did Spain’s Pedro Sánchez and Norway’s government. The European Union’s collective statement, issued through High Representative Kaja Kallas, called only for “full respect of international law,” language abstract enough to avoid specifying whose conduct had violated it. Germany’s chancellor Friedrich Merz went further in the wrong direction, suggesting publicly that international law should not stand in the way of necessary action against a regime that disregarded the law itself, a justification uncomfortably familiar from 2003. French and German leaders found the spine to oppose the Iraq invasion outright that year. In 2026, the same governments mostly found vague language and quiet base access instead. Analysts describe this as reflexive obedience rather than strategic calculation, rooted in European anxiety about preserving American backing for Ukraine. Ironically, a prolonged Iran war diverts the very air-defense systems and political attention Ukraine needs, making the bargain self-defeating.
Sovereignty anxiety triggered by the Anthropic cutoff deserves the same scrutiny, because its use so far has little to do with restraint. The headline European response has not been a serious push to bring military AI inside the AI Act, or to build independent oversight of algorithmic targeting, or even to pause procurement of systems like Maven pending an accounting of Minab. Instead it has fed a defense-spending boom, the roughly 800 billion euro ReArm Europe package, justified explicitly in the language of AI sovereignty. Some of that money will likely flow toward the same companies, Palantir prominent among them, now profiting from systems under scrutiny for Minab. One commentator put it starkly to Common Dreams: an arms industry effectively going public off the back of a war that killed children at a school. A continent genuinely afraid of unaccountable algorithmic warfare would race to write enforceable rules before building its own version of the system. A continent managing a sovereignty panic races to buy one instead.
A fair objection deserves a hearing rather than dismissal. European governments genuinely lack the leverage to halt a war Washington launched without consultation, and Iran’s advancing nuclear program gave even skeptical governments real proliferation concerns to weigh, not just cowardice to indulge. Some of the disjointedness reflects honest disagreement among twenty-seven sovereign states rather than bad faith. These are real constraints, though, not justifications for refusing to call an unlawful war by its name, or for treating new dependency on American algorithms mainly as a procurement opportunity. The honest accounting runs as follows: The war on Iran did change Europe’s perception of AI in warfare, from a future risk into a demonstrated present one, and the change is real and overdue. Perception is not policy, and alarm is not accountability. European leaders willing to call the Anthropic cutoff a sovereignty crisis need to apply equal clarity to the strike on a girls’ school as a consequence of a war fought in violation of the UN Charter. Money flowing toward European AI independence needs enforceable limits attached, not just fresh contracts for the firms that built the original kill chain. Until both happen, Europe’s awakening amounts to little more than learning to fear the algorithm while still declining to confront the war that revealed exactly what it can do.
Further Reading on E-International Relations
