G7 Leaders Discuss Trusted Access to Advanced US AI
Leaders at the G7 summit in Evian-les-Bains weighed a practical — and politically delicate — workaround to recent U.S. restrictions on cutting‑edge artificial intelligence. Diplomatic sources said officials discussed a plan to give a small group of “trusted partners” controlled access to advanced models developed by U.S. companies like Anthropic, a move that could let allies benefit from the technology while keeping national security concerns in check.
The discussion followed a dramatic step by Anthropic last week, when the company disabled access to its most capable models, Fable 5 and Mythos 5, for all users. The decision came after President Donald Trump ordered Anthropic to block foreign nationals from using those systems, citing fears that the powerful tools could be abused or used against U.S. interests. The result was abrupt: researchers, companies and governments outside the United States suddenly found themselves cut off from some of the most advanced AI systems available.
Diplomats and officials at Evian were trying to square two competing priorities: preventing potentially dangerous misuse of next‑generation AI, and ensuring allies can still harness those tools for defense, cybersecurity and economic resilience. One focus of the talks was a mechanism to offer “trusted” G7 partners conditional access under strict safeguards — essentially a managed pipeline that would let allies use advanced models for legitimate security and research work without exposing them to unfettered foreign use.
Behind the idea is a straightforward rationale. Some of these models, particularly Anthropic’s Mythos, are explicitly designed to find vulnerabilities in software and identify weaknesses in code. That capability can be used defensively — to harden financial systems, secure infrastructure and test critical networks — but it could also accelerate offensive cyberattacks if it fell into the wrong hands. Cybersecurity officials at the summit argued that allowing vetted allies to study and deploy such tools could strengthen collective defenses against rival actors, notably China, which many G7 countries view as a strategic competitor in both cyberspace and AI development.
U.S. Commerce Secretary Howard Lutnick was closely involved in the discussions on the sidelines of the summit dinner, according to sources, reflecting Washington’s interest in balancing technology control with alliance management. European officials, for example, have been particularly eager to gain access to Mythos to assess its implications and develop regulations or defensive responses. The European Union has already signaled a desire to study the model in order to understand potential threats and craft policy responses.
The talks in Evian illustrate the awkward new phase of international tech diplomacy: advanced AI is now as much a geopolitical asset as a commercial product. Nations want the benefits — whether faster vulnerability scanning, improved cybersecurity, or economic competitiveness — but are wary of the risks of wider proliferation. Finding a system of shared, managed access would require detailed agreements on oversight, data handling, export controls and technical safeguards to prevent misuse.
Whether the G7 will formalize such a mechanism remains uncertain. The conversation, however, signals a shared recognition that allies cannot simply be shut out of vital technologies without consequences. The challenge now is to design practical guardrails that preserve security without strangling collaboration — a delicate balancing act that will test both policymakers and the companies that build the systems.