A quiet but consequential debate is emerging at the G7 summit over who should be allowed to work with the most advanced United States artificial intelligence models and on what terms. Behind closed doors, leaders are testing a formula that would open a narrow channel of access for selected allies while preserving Washington’s new controls on foreign use of cutting edge systems.
Trusted Partners as A New Access Club
According to several diplomatic sources, G7 leaders discussed a scheme under which a limited group of trusted partners would be granted access to United States frontier models developed by firms such as Anthropic. These partners could include both governments and companies, effectively creating a tiered system of access to high end models that are currently restricted for non United States nationals. The conversation reportedly unfolded on the margins of the opening summit dinner in Evian les Bains with United States officials, including Commerce Secretary Howard Lutnick, sounding out allied views.
This discussion directly responds to Washington’s recent decision to suspend access to Anthropics most advanced models for foreign users on national security grounds. The Trump administration previously ordered Anthropic to withdraw its newest models from overseas deployment, a move described by Canadian Prime Minister Mark Carney as a warning about dependence on a small number of United States providers. The proposed trusted partner channel would therefore operate less as a liberalisation than as a calibrated exception to a tightening export control environment.
Strategic Controls or Fragmented AI Order
The emerging approach fits a broader trend in which the G7 has treated advanced AI as a strategic technology to be governed through coordinated safeguards and export constraints. Previous G7 processes on trustworthy and generative AI highlighted safety, security and resilience, and have signalled that high risk systems may require extra oversight and cross border alignment. Yet a unilateral restriction followed by a selective exemption mechanism could accelerate the fragmentation of the global AI landscape into access clubs anchored around a few major powers.
For close allies, inclusion as trusted partners would secure early access to frontier models and associated research benefits, but it would likely come with stringent conditions on data handling, model use, and potential onward sharing. Those outside the circle could face higher costs, slower capability development, or pressure to align politically and technologically with alternative AI ecosystems. The G7s stated ambition to promote inclusive and equitable AI for sustainable development sits uneasily with a system that differentially gates access to the most capable models.
Balancing Security Fears and Dependence Risks
United States officials frame tighter controls as necessary to mitigate risks that frontier models could be misused in domains such as cyber operations or biotechnology if accessed by adversaries or poorly regulated actors. By carving out a trusted group, Washington appears to be seeking a compromise that reassures allies while keeping ultimate discretion over who can work with the most powerful systems. However, partners already express unease that sudden restrictions can disrupt research agendas and industrial strategies built around access to United States technology.
Carney has argued that the recent United States measures underscore the need for countries to diversify their AI supply base rather than rely on a handful of American firms. If a trusted partner scheme is perceived as unpredictable or politically contingent, it may push other states to invest more aggressively in domestic or alternative AI capabilities to reduce vulnerability. That dynamic could, over time, reduce the very leverage that export controls and access clubs are designed to preserve.
Implications for Global AI Governance
Beyond immediate questions of access, the Evian discussions signal that frontier model governance is moving from principles to concrete allocation mechanisms. G7 statements in recent years have emphasised safety standards, transparency and cooperation with developing countries, including plans under the Verona process to support AI hubs and open, secure data models for sustainable development. A restricted trusted partner regime risks creating a two track order in which high capacity models circulate among a narrow group while other states are encouraged to work with less advanced or more open systems.
This tension matters because G7 coordination often shapes broader multilateral norms, from export controls to regulatory benchmarks. If access decisions hinge on security assessments that are not fully transparent, it may be harder to persuade non G7 countries that AI governance frameworks are genuinely inclusive and rules based. At the same time, if the G7 can embed clear criteria, oversight mechanisms, and pathways for broader participation, the trusted partner idea could evolve into a structured tool for risk management rather than a purely geopolitical filter.
A Final Note
The debate over trusted access to United States frontier AI models illustrates how security, economic strategy and governance ambitions are converging around a few highly capable systems. Whether the G7 can translate this ad hoc discussion into a transparent, balanced framework will shape not only allied cooperation but also the credibility of its wider agenda on safe and inclusive AI development.
