G7 leaders consider a pathway for 'trusted partners' to access America's leading AI models.

      The exchange took place in a typical setting for such discussions, during a dinner. At the start of the G7 summit in Evian-les-Bains, France, on Monday, representatives from various member countries brought up with US Commerce Secretary Howard Lutnick the concept of a ‘trusted partners’ initiative that would allow allied nations or companies access to the most advanced American AI models, according to sources familiar with the talks.

      This proposal is a direct response to a decision made in Washington just four days prior. On June 12, following a directive from the US Commerce Department, Anthropic restricted access to its two most advanced systems, Mythos 5 and Fable 5, to foreign nationals. Since this restriction couldn’t be selectively enforced on a shared cloud platform, the company opted to shut down the models for everyone globally. This order is thought to be the first export-control measure targeting specific AI models rather than the chips powering them.

      This new approach has particularly concerned the allies. They are accustomed to chip controls, as the US has spent years limiting the sale of hardware and dictating to whom it can be sold. However, denying access to the models themselves, even to citizens of close partners, represents a more recent and sharper tool, and the G7 representatives dedicated their dinner on Monday to seeking a solution to a newly shut door. The ‘trusted partners’ concept under consideration could include either countries or individual corporations, the sources indicated.

      The stated justification for this is based more on security rather than commercial interests. The argument posits that broader access to cutting-edge models would enhance G7 countries' cyber defenses against adversaries, particularly China. This appeal aligns with Washington’s own rhetoric: provide allies with the best tools, and they will perform better in areas of mutual concern. Whether this framing will influence the administration remains uncertain.

      For Anthropic, this incident marks another chapter in its complex relationship with the US government. The Pentagon has labeled the company as a supply-chain risk, federal agencies have been directed to cease using its technology, and it has even taken legal action against the government over the blacklisting, while simultaneously receiving instructions from other government branches for banks to utilize its models.

      The June directive to deactivate its most advanced systems is the most significant intervention to date and has now brought the G7 into the discourse.

      Nothing has been finalized; the conversations mentioned were exploratory, taking place between representatives of countries and US officials rather than being formally endorsed by leaders. Consequently, a ‘trusted partners’ arrangement currently remains a proposed idea discussed over dinner.

      What is clearer than the potential outcomes of these discussions is that America’s closest allies have realized that the premier AI they can access is American, that this access has become a tool for Washington to wield, and that their only recourse is to politely request it back.