At the G7 summit, France and India asked who holds the off switch on American AI. The U.S. justification for the Anthropic block suggests that lever was already half loose.
The U.S. blocked Anthropic's newest AI models from export last week on national security grounds. The capabilities the government cited as the reason are already freely available in OpenAI's models, cybersecurity experts told TechCrunch. That gap sat at the center of an awkward G7 lunch in Canada on Wednesday, when France and India pressed Washington on the price of allied dependence on American artificial intelligence.
French President Emmanuel Macron and Indian Prime Minister Narendra Modi raised the concern directly with G7 leaders, the chief executives of Anthropic and OpenAI, and President Donald Trump, per TechCrunch's reporting on the summit. Macron warned that if the U.S. could, "from one day to the next...turn off the switch," allied customers and American AI firms would both pay the price. Modi pressed for sovereign guarantees. The shared argument was that the world's democracies cannot plan around AI infrastructure they cannot keep.
The trigger for the policy fight was a flag from Amazon. The company told the White House that certain safety guardrails on Anthropic's Mythos 5 and Fable 5 models could be bypassed, according to TechCrunch. Within days, the Trump administration imposed a national-security export block on both systems. Cybersecurity experts have argued that the underlying capabilities were not unique to Anthropic: the same functions are present in models that remain freely available, including those from OpenAI.
Read narrowly, the export block is a safety move against two specific Anthropic systems. Read against the cybersecurity experts' claim, it is a policy that pulls one American frontier lab's models out of allied reach while leaving comparable tools one download away. The U.S. government can defend that distinction in technical terms, but it is hard to defend it as a capability ceiling. The lever it is trying to keep is, on its own account, already half-loose.
The corporate intermediary matters here. Amazon's flag turned a technical safety concern into the legal predicate for an export block in days. That speed, and the asymmetry of which American companies can move the White House that fast, recasts the story from pure geopolitics to industrial policy in a safety wrapper. Anthropic, OpenAI, and the allied governments now negotiating access all have to ask whether future U.S. action turns on engineering evidence, or on which American cloud company is loudest in the room.
Aidan Gomez, co-founder and CEO of Cohere, framed the allied position directly in a statement shared with TechCrunch: "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."
The next two documents to watch are the Commerce Department's Bureau of Industry and Security public notice on the Mythos 5 and Fable 5 restrictions, and Anthropic's formal response. The notice will tell allied governments how narrow or how wide the line is, and whether Washington is drawing it on capabilities or on companies. Anthropic's response will tell customers and competitors whether a frontier model can survive being cut off from its largest export markets, and what kind of sovereign insurance, if any, allied governments can realistically demand.