The G7 AI Summit Buried Multilateral Governance Without Saying So

Ahead of the G7 AI summit in Évian-les-Bains, two actions by the Trump administration demonstrated that the United States controls not only advanced AI technology but also the conditions under which others may access it. The responses of leading AI executives reflected accommodation rather than resistance, reinforcing the assumption of continued American leadership. Despite possessing regulatory and industrial tools, Europe remains focused on managing dependence rather than achieving technological sovereignty. As a result, Europe’s strategic vulnerability in AI stems as much from political choice as from technological constraints.

Two events preceded the G7 AI summit in Évian-les-Bains, and neither was on the agenda. In February, the Pentagon terminated its $200 million classified contract with Anthropic, designating it a national security supply chain risk; the first time this was applied to an American firm. The trigger was Anthropic's refusal to remove restrictions on mass domestic surveillance and autonomous weapons from its acceptable use policy. The message: compliance is the price of market access in the Trump era. Days before the summit, the US Commerce Department blocked all foreign nationals, including Anthropic's own employees, from accessing Fable 5 and Mythos 5. The procurement weapon and the model switch-off were a single demonstration that the US controls not just the tech, but the terms of global access.

On June 17, three CEOs sat down with Trump and Macron: Amodei (Anthropic), Altman (OpenAI), and Hassabis (DeepMind). Their subsequent public statements revealed three accommodation strategies, none constituting resistance. Amodei accepted the security rationale, proposing to expand the sovereignty perimeter to democratic allies, conditional alignment, not opposition. Macron warned that nobody will buy American AI if they fear it can be switched off, not a call for independent EU governance, but a warning about losing access. Altman called for an international safety forum, leaving American dominance intact. One common premise: the US leads. The only variable left is the shade of grey within that reality.

Two counter-levers went unplayed. Macron did not play the Mistral card despite its open-weight models, EU data residency, and a significant investment in French infrastructure. Invoking Mistral would have exposed Europe's infrastructure gap rather than closing it. Amodei could have threatened to relocate strategic operations to Europe’s deep talent base, but did not: Amazon, a major investor, reportedly helped trigger the export control. Relocation meant corporate self-destruction. Frame internalization is complete when actors cannot assemble credible counter-leverage even in private.

Europe’s remaining tools, restricted procurement, fiscal incentives, and deploying the AI Act as a structural preference, are dependency management tools, not sovereignty strategies. This is the defining structural fact of European AI policy in 2026, prioritizing integration over independence. The calculation is simple: better a belligerent technology that works than a non-belligerent one that may not. This is a precise asymmetric risk calculation under industrial time pressure. The kill switch works not because Washington built it, but because Europe repeatedly chose to remain within its range.

John S. Wyse is a researcher in cognitive linguistics and strategic communication, Department Director at the European School of Economics, and a professor at GEMA Business School. His research focuses on the impact of figurative language and cognitive frames on highly strategic decision-making processes, with a particular emphasis on the role of communication in international politics, leadership, and negotiation.

Photo: PM in a family photo with World leaders on the sidelines of the G7 Summit, in Evian, France on June 16, 2026. By Prime Minister's Office (GODL-India), GODL-India, https://commons.wikimedia.org/w/index.php?curid=194007434