On June 12, the Trump administration restricted Anthropic’s two most powerful models, Fable 5 and Mythos 5. It invoked the 2018 Export Control Reform Act, citing national security concerns tied to their cyber capabilities. Because Anthropic couldn’t verify the nationality of every user, it had to disable both models worldwide. That cut off everyone, including its own non-US employees. Talks with the Commerce Department ended Monday with no resolution.
Five days later, at a closed-door G7 lunch in Évian, France, two of those CEOs made a striking ask. Anthropic’s Dario Amodei and Google DeepMind’s Demis Hassabis called for a US-led coalition to set global AI rules. Sam Altman called for an international testing forum. Canada’s Mark Carney said the US could lead. It was the first face-to-face between Amodei and Trump since the ban. Trump said the negotiations were “going fine.”
The contradiction is the story. The AI labs want the US to lead the world on AI governance. At the same time, they’re fighting the US over its unilateral power to switch their products off. And the ban itself rattled every American ally in the room. It proved their economies could come to depend on technology Washington can revoke without warning.
Best for anyone tracking AI policy, the Mythos saga, or the collision between frontier labs and governments. Not ideal for readers who want a tidy resolution, because there isn’t one yet.
The Two Sentences That Tell the Whole Story
On Friday, the most powerful AI models Anthropic has ever built went dark. The Anthropic export controls had landed, and the company’s flagship models went with them.
Not because of a bug. Because the United States government flipped a switch.
Five days later, the CEO whose flagship products had just been forced offline stood in front of Donald Trump and the leaders of the G7. And he asked the United States to lead the entire world on AI.
Read those two sentences again. That’s the whole story.
What the Anthropic Export Controls Actually Did
On June 12, the Trump administration imposed export controls on Anthropic’s Fable 5 and Mythos 5, the company’s two most capable models. The order used the 2018 Export Control Reform Act, marking the first time the government has regulated AI as an emerging technology under that law. The Anthropic export controls cited national security as the stated reason. These models are unusually good at finding software vulnerabilities. The government’s fear is straightforward. A tool that good at finding holes is a tool that good at exploiting them.
Here, the mechanism is the brutal part. The controls block all non-US citizens from accessing the models. Anthropic had no way to verify the nationality of every user on its platform. So to comply, it didn’t restrict the models. It shut them down entirely, for everyone, worldwide. Anthropic’s own employees who aren’t US citizens lost access to the company’s flagship product. In short, a frontier lab had to pull its best work offline because it couldn’t prove who was using it.
Meanwhile, Anthropic dispatched senior officials to negotiate with the Commerce Department. Those talks wrapped Monday with no resolution. As of the G7 summit, the Anthropic export controls were still firmly in place. This is the backdrop everyone in Évian was operating against.
What Happened at the G7
On Wednesday, on the final day of the summit in Évian-les-Bains, about a dozen tech executives sat down with the heads of the G7 nations. Notably, Amodei was there. So were OpenAI’s Sam Altman, DeepMind’s Demis Hassabis, and Mistral’s Arthur Mensch. Trump was there too, flanked by Treasury Secretary Bessent, Commerce Secretary Lutnick, and Secretary of State Rubio.
And in that room, with the man who had just banned his flagship models sitting nearby, Amodei nonetheless made his pitch. He called for a US-led coalition to shape international AI rules and standards. He argued that cooperation should include structured access to frontier models and trade in chips and critical components that excludes China, a pitch The Next Web noted sat awkwardly against the Anthropic export controls still in force. Hassabis backed a US-led approach and pushed for an international standard-setting body. Altman called for a global forum to test and verify AI capabilities. Canadian Prime Minister Mark Carney reportedly agreed the US could lead such a coalition.
Still, no binding commitments came out of it. The expected outcome was always voluntary pledges, not enforceable rules. Trump, seated next to Altman, didn’t bring up the Anthropic ban at all during the lunch, according to a person familiar with the meeting. Afterward he told reporters the negotiations with Anthropic were “going fine” and pivoted to talking about building power plants.
The Contradiction Nobody in the Room Said Out Loud
Meanwhile, here’s the thing the polite coverage will dance around.
Amodei asked the United States to lead global AI governance five days after the United States forced his company’s products offline. Those two positions are in direct tension, and the tension is the actual news.
After all, the labs have spent two years positioning themselves as indispensable partners to Washington. They show up at summits. They brief the White House. They argue that frontier AI in the hands of defenders beats frontier AI in the hands of attackers. That’s the exact case Anthropic made for Mythos, before the government pulled it anyway. And now they’re asking that same government to set the rules for everyone. Meanwhile they’re privately fighting its use of the bluntest rule it has: the power to switch a model off.
Ultimately, you can read this two ways, and both are true. The charitable read is that Amodei genuinely believes US-led governance is the least-bad option. A coordinated framework beats the unilateral, unpredictable bans he just got hit with. Asking for structure is a way of asking for rules that aren’t decided by one administration’s mood on a Friday afternoon.
The cynical read is that the labs want it both ways. They want the legitimacy and market access that comes from being America’s chosen AI champions, and they want to avoid the control that comes with it. “Lead the world, but don’t switch us off” is not a stable position. It’s the sound of an industry discovering that being a strategic national asset means being treated like one.
Why the Allies Are Spooked
The part of this that matters beyond Anthropic is what it did to everyone watching.
The US shut down a company’s product with little public explanation. And it did so in a way that reached across borders, pulling the model out of every other country’s hands too. Per the Washington Post’s reporting, the Anthropic export controls alarmed the global leaders at the summit. They now fear their economies could come to depend on technology Washington can withdraw without warning.
Indeed, that fear has a name, and it’s been building all year: sovereign AI. Multiple G7 nations had already talked about investing in their own AI capacity. But the assumption was always that this would happen alongside access to the US tech stack, not instead of it. The Mythos ban broke that assumption. As one Atlantic Council fellow put it, the export controls “changed everything.” If the US will cut off its closest allies from a frontier model, then every nation that isn’t the US has a new reason to build its own.
For instance, France, hosting the summit, is already pushing this line through Mistral. The EU has its own regulatory path that doesn’t presume American leadership. So when Amodei asked for a US-led coalition, he was asking a room that had just watched the US demonstrate exactly why they might not want to depend on it. The timing could not have undercut the pitch more cleanly.
Where This Goes
The Mythos story has been a slow-motion collision between a frontier lab and the realities of what it built. This is the moment the government became the main character. We’ve tracked it from the original Mythos leak and the asymmetry it created to Google naming OpenClaw in a hacker threat report, and the throughline has always been the same. These models are good enough at offense that nobody, not the labs, not the governments, has a clean answer for who should hold them.
So, three things to watch. First, whether the Commerce Department talks produce any path to restore Fable 5 and Mythos 5, or whether they stay dark indefinitely. A frontier lab with its best models offline is a frontier lab bleeding its lead every day the ban holds. Second, whether any allied nation actually accelerates sovereign-AI spending in direct response. France and the EU are the ones to watch. Third, whether the “US-led coalition” idea survives a simple fact. The US just demonstrated it will act alone whenever it decides national security requires it.
Ultimately, the labs spent two years telling us they were building something powerful enough to change the world. The US government just agreed with them, and treated Mythos like a weapon instead of a product. The labs wanted to be taken seriously. They got their wish. Now they’re learning what it costs to be the most important thing in the room. They’re asking the people who can switch them off to please, carefully, write the rules.
