The G7 summit in the French Alpine resort of Evian-les-Bains was meant to showcase democratic coordination on artificial intelligence. Instead, it was overshadowed by a unilateral American decision that sent shockwaves through allied capitals. Days before the leaders convened, the Trump administration ordered Anthropic to bar foreign nationals from accessing its most advanced AI systems — Mythos 5 and Claude Fable 5 — citing national security concerns. Because cloud environments cannot reliably distinguish users by nationality in real time, the California-based firm was forced to take both models offline globally, abruptly severing access for governments, research institutes, and corporations across Europe, Canada, and Asia. The move, viewed from Washington as a necessary safeguard against misuse, was received in Paris, Ottawa, and Brussels as a stark demonstration of how a single country can unilaterally pull the plug on critical technology.

Against this backdrop, OpenAI chief executive Sam Altman delivered an unusually self-effacing appeal. “Do not cede your responsibilities to AI labs like mine,” he told the assembled heads of state and government, urging them to set global standards for deploying the rapidly advancing technology. His intervention, while welcomed by many, also underscored the asymmetry at the heart of the crisis: the very companies building frontier AI are asking democratic governments to reclaim authority that has, in practice, already slipped from their grasp. French President Emmanuel Macron seized the moment to advance a “trusted partners” scheme that would grant vetted non-US nations structured access to American models, holding separate discussions with Altman and Anthropic’s chief executive Dario Amodei on the summit’s sidelines.

The trigger for Washington’s clampdown, according to officials cited in American press reports, was the expansion of Anthropic’s early-access programme. A list of roughly 150 organisations across more than 15 countries included a South Korean telecom operator that White House reviewers suspected of maintaining links to China. Although SK Telecom, widely identified as the firm in question, has denied any such connection and stressed it does not use Huawei equipment in its core network, the episode eroded confidence in Anthropic’s ability to safeguard sensitive capabilities. The ban thus swept up not only the Korean company but also major European and Canadian institutions that had been granted access in good faith.

European lawmakers reacted with alarm. A French member of the European Parliament, Christophe Grudler, described the episode as proof that the United States holds a “kill-switch” over essential digital infrastructure. His Dutch colleague Bart Groothuis warned that without homegrown large language models and open-weight alternatives, Europe risks a form of “digital colonisation.” Canada’s Prime Minister Mark Carney struck a more measured tone, insisting that “nobody did anything wrong,” yet urged countries to learn the lesson and diversify their technology sources. France has already announced plans to shift government departments towards domestic AI and data analytics tools, reducing reliance on American cloud services — a move that analysts in London see as the first concrete step in a broader European pivot.

Viewed from Brasília, the crisis looks different. Brazil’s director of information security, Danielle Ayres, said the country is monitoring the advance of AI systems capable of identifying network vulnerabilities but does not consider itself in a position of fragility. The principal challenge, she noted, is keeping pace with the speed of technological change. That stance reflects a wider reality: while the immediate shock has been felt most acutely by close US allies, the longer-term consequences will ripple across the entire global AI supply chain. The Anthropic affair may well be remembered as the moment when the governance of artificial intelligence ceased to be a technical debate and became, irreversibly, a contest of geopolitical power.