G7 Explores 'Trusted Partner' Access to U.S. Advanced AI Models

Leaders from the Group of Seven nations are exploring a framework that would grant select "trusted partners" access to cutting-edge AI models from U.S. companies like Anthropic, according to three diplomatic sources who spoke about discussions at this week's summit in France.

The conversations follow Anthropic's Friday decision to disable global access to Fable 5 and Mythos 5, its most advanced AI systems, after President Donald Trump directed the company to block foreign nationals from using these models on national security grounds.

Sidebar Diplomacy at Evian Summit

Multiple country representatives raised the access issue with U.S. officials during the G7 gathering in Evian-les-Bains, particularly in discussions with Commerce Secretary Howard Lutnick on the sidelines of Monday's opening dinner. The potential "trusted partners" designation could apply to both nations and individual companies, according to a second diplomatic source who requested anonymity due to ongoing negotiations.

A Trump administration official confirmed the White House maintains "an open line of communication with our allies" and remains focused on addressing national security concerns related to Anthropic's technology.

Why It Matters

The diplomatic push reflects tension between AI export controls and alliance relationships. G7 nations argue they need access to the most advanced models to build robust cybersecurity defenses against adversaries like China. Without a trusted partner framework, U.S. restrictions could leave allies vulnerable while concentrating AI capabilities domestically—a dynamic that complicates coordinated responses to shared security threats.

Critical Infrastructure at Stake

The access question carries particular weight for Anthropic's Mythos model, which is specifically designed to identify vulnerabilities in computer code. Cybersecurity experts warn the system could accelerate attacks on banking infrastructure if misused, yet the European Union is actively seeking access to study the model's security implications.

Before Trump's directive, Anthropic had granted Mythos access to select organizations across more than 15 countries, including entities in healthcare, communications, power, and water sectors that used the tool to scan their systems for weaknesses.

AI executives from Anthropic, OpenAI, and Google are scheduled to attend a working lunch Wednesday at the summit to discuss technology policy issues including regulation, AI infrastructure, and network security.

Anthropichas not responded to requests for comment on the trusted partner discussions.

These details were first reported by Reuters.