G7 leaders pledge closer AI coordination, eye 'trusted partners' scheme to bypass US export restrictions

The final day of the G7 summit in France brought artificial intelligence to the centre of the diplomatic agenda, with heads of state and top tech CEOs grappling with the fallout from Washington's sudden export ban on Anthropic's most powerful models.

US export controls rattle allies

On 12 June, the US Commerce Department ordered Anthropic to suspend foreign access to its Fable 5 and Mythos 5 models, citing national security after a reported jailbreak. Because the firm could not quickly build nationality-based controls, it disabled the models globally, cutting off Americans and foreigners alike. The move alarmed allied governments, who fear their critical infrastructure and economies could become dependent on a technology that can be switched off overnight.

“... the very strong decision from the Trump administration is also a bad thing... the reaction is in some regards strictly nationalist.”

'Trusted partners' scheme proposed

Against that backdrop, G7 leaders discussed a "trusted partners" mechanism that would let approved countries or companies access advanced US AI models, creating a path around the restrictions. The goal is to maintain an open trade network among democracies while tightly controlling which actors can use the technology. India's prime minister also voiced concern, arguing that democratic nations need unfettered access to top AI models to protect critical systems.

“We cannot rely on tools developed by foreign powers. France must have its own tools.”

Europe pushes for tech sovereignty

The ban intensified Europe's long-running push for digital independence. EU officials pointed to plans for AI "gigafactories" and new laws to boost domestic cloud, chip and AI industries. French and British ministers said the episode proved the danger of overreliance on a handful of US labs. The UK spent the weekend lobbying Washington for a carve-out but was told there was "zero chance," leaving thousands of British businesses without access.

“Everybody knows that it's really important that G7 countries and its allies band together — that we win in the AI race and.. don't let autocratic countries take the winning position.”

US-led AI coalition floated, no binding deal

Anthropic's Dario Amodei and DeepMind's Demis Hassabis pitched a US-led coalition to set international AI rules, coordinate chip trade, and exclude China. Canadian Prime Minister Mark Carney agreed the US could lead such a body. OpenAI's Sam Altman called for a global testing forum. But the working lunch produced no binding commitments, only a joint statement pledging to assess frontier AI's impact on financial stability and labour markets.

“The recent restriction on access to Anthropic's models confirms what we at Cohere have known all along: that companies and democratic nations remaining dependent on a small handful of big tech companies is dangerous to resilience.”