← Back to feed

The Trump administration’s June 13 export control directive arrived without warning. Anthropic received an order to suspend foreign access to Fable 5 and Mythos 5 and took both models offline globally to ensure compliance. Two hundred institutions across 15 countries lost access the same day — including governments and research bodies Anthropic had enrolled in its Mythos Preview vulnerability-testing program.

The incident dominated the AI track at the G7 summit held this week in France.

What G7 Leaders Said

French President Emmanuel Macron addressed the summit with Anthropic CEO Dario Amodei, OpenAI CEO Sam Altman, and Google DeepMind CEO Demis Hassabis present. His framing was direct: if the United States can “from one day to the next turn off the switch,” it does not just damage European economies — it damages the American AI companies whose commercial future depends on international trust.

Indian Prime Minister Narendra Modi echoed the concern with harder language, arguing that democratic nations must have guaranteed access to frontier AI to protect critical information infrastructure. The implicit argument: AI is now infrastructure, and infrastructure you don’t control can be used as leverage against you.

All three AI CEOs expressed support for some form of international cooperation. Altman told leaders not to “cede your responsibilities to AI labs like mine” — a framing that acknowledged the governance gap without offering a concrete mechanism. Amodei argued for coordination on model access, semiconductor supply chains, and biosecurity and cybersecurity risks across allied nations.

The Trusted Partners Proposal

G7 leaders formally discussed a “trusted partners” scheme. Under the proposal, allied nations would receive assured access to advanced models from US firms including Anthropic and OpenAI, contingent on using those systems to strengthen defences against adversaries — primarily China. Macron backed the framework as a path around US restrictions that could also restore customer confidence in American AI products.

No binding commitments were made. The joint G7 statement committed to examining how frontier AI affects economic productivity, labor markets, financial stability, and cybersecurity. It did not announce timelines, enforcement mechanisms, or a formal working group.

The Gap the Summit Didn’t Close

The structural problem with a trusted partners scheme is that it requires the same White House that issued the original export order to guarantee the override. A diplomatic framework negotiated with the current administration can be rescinded by executive action with the same speed as the order that triggered the discussion.

The June 13 order itself appeared to originate from Amazon flagging potential guardrail vulnerabilities in Mythos 5 to the White House — a private-sector risk report that led to a government export control within days. The speed from flag to cutoff suggests the access the scheme is trying to protect can be revoked faster than any diplomatic process can protect it.

France’s parallel track is more concrete. Mistral, the only European lab with a plausible claim to frontier capability, received accelerated government investment ahead of the summit. The EU’s May tech sovereignty package, which would bar US cloud providers from government health and finance data, is moving through Brussels. G7 allies are pursuing the diplomatic option and the independence option simultaneously.

Key Numbers

• Institutions affected: 200 across 15 countries
• Warning period before cutoff: Zero
• Models taken offline: Fable 5, Mythos 5
• Trigger: Amazon flagged guardrail vulnerabilities to White House
• Mythos Preview program enrollees: 200 institutions testing for vulnerabilities
• Trusted partners proposal: Allied-nation access contingent on using models against shared adversaries
• Binding commitments from summit: None announced