G7 Weighs a 'Trusted Partners' Framework for Access to Restricted US AI Models

G7 Weighs a 'Trusted Partners' Framework for Access to Restricted US AI Models

Days before the leaders of the world's wealthiest democracies sat down to lunch in the French spa town of Evian, the United States did something it had never done before: it pulled the plug on a frontier AI model. A Commerce Department export directive forced Anthropic to disable its most advanced systems, Fable 5 and Mythos 5, for every user on Earth — not just the foreign nationals the order targeted, but American customers too. By the time President Donald Trump and his counterparts gathered on June 17, 2026, the question hanging over the summit was no longer abstract. Allied governments had just watched Washington reach across borders and switch off software their own economies were starting to depend on. The diplomatic scramble that followed produced the summit's most consequential AI idea: a "trusted partners" framework.

The proposal, raised by representatives of several G7 nations with U.S. Commerce Secretary Howard Lutnick on the sidelines at Evian, would create a tiered system of vetted allies — countries or companies cleared, like a security clearance, to access restricted American frontier models. According to G7 diplomatic sources cited by Reuters, no agreement was reached. What emerged was a negotiation, not a communique line: allied access to the most capable AI models is now being haggled over as a matter of trade and national security, the way enriched uranium or advanced semiconductors long have been.

From export control to bargaining chip

The trigger was the "deemed export" doctrine — the long-standing rule that giving a foreign national access to controlled technology counts as an export, even if no goods cross a border. Applied to a cloud-hosted AI model, it gave Washington what European officials had quietly feared for years: an off switch. Reporting indicates the Anthropic shutdown cascaded from a single safety flag, and the practical lesson for allies was blunt — any economy built on U.S. AI infrastructure now holds a dependency it does not control.

Inside the closed-door working lunch, the tech industry pushed the partners idea hard. Anthropic CEO Dario Amodei and Google DeepMind's Demis Hassabis, joined by OpenAI's Sam Altman and roughly a dozen other executives, called for a U.S.-led AI coalition to set shared rules. Amodei told the group that areas of international cooperation should include structured access to frontier models for trusted partners, plus coordinated trade in chips and critical components that explicitly excludes China. The pitch reframes export controls not as a wall but as a gate: keep adversaries out, let allies in, and use the combined firepower to build cybersecurity defenses against Beijing.

For the vendors, the logic is also commercial. A model that allies cannot legally touch is a model that cannot be sold abroad, and a trusted-partners regime would restore market access while letting Washington keep its grip on who ultimately qualifies.

The sovereignty backlash

Not everyone at Evian read the framework as reassurance. French President Emmanuel Macron made what he described as a forceful plea for the U.S. not to keep cutting-edge AI to itself. He allowed that it was a "good thing" American officials recognized frontier models could be dangerous, but called the export crackdown a "strictly nationalist" reaction. "So let us move forward together," Macron said. "Our relevant agencies must first cooperate so that, in the areas of security and cybersecurity, we have a smooth government-to-government relationship." India's Narendra Modi pressed the same uncomfortable point directly: what happens to a national economy when the AI models it runs on can be turned off by a foreign capital?

Macron's answer was institutional. He announced that Western democracies would stand up a coordinated AI cooperation platform within a month, with a follow-up leaders' meeting set for September. The framing matters. A trusted-partners scheme, as floated by U.S. officials and vendors, keeps the controlling hand in Washington — allies receive access on American terms. Macron's platform points the other way, toward a shared European and democratic effort to reduce dependence rather than formalize it. The two visions are not the same, and the gap between them is the real story.

That tension exposes the central paradox of frontier AI diplomacy in 2026. Export controls are designed to deny capability to rivals, but the most capable models are now general-purpose infrastructure that allies want to build on, not merely buy. A "trusted partner" who can be untrusted by directive is, from a sovereignty standpoint, not sovereign at all. China, watching from outside the room, has begun pitching free AI to the developing world — a reminder that every model Washington withholds is an opening for a competitor that asks fewer questions.

What to watch

The immediate test is whether the trusted-partners concept survives contact with a U.S. administration that just demonstrated its willingness to switch a model off. Watch for any tiering details — which countries land in the top tier, whether the United Kingdom's lapsed exemption is restored, and whether "partners" means governments, companies, or both. Watch Macron's one-month deadline and the September leaders' meeting: a real platform with shared compute and standards would signal that allies are hedging, not waiting. And watch Anthropic, whose models sit at the center of the dispute and whose CEO is now openly lobbying for the very access regime that would govern them. The frontier-model era has produced its first true diplomatic crisis. How the G7 resolves who gets the keys — and who keeps the right to take them back — will set the terms for the next one.