On Friday night, the Commerce Department ordered Anthropic to restrict access to Fable 5 and Mythos 5 for foreign nationals—including the company's own employees working in the US. The trigger? A jailbreak vulnerability reported by Amazon researchers. The outcome: the most capable AI models on the planet are now US-persons-only.
What Actually Happened
Anthropic released its Mythos-class models under the commercial name Fable earlier this week. Within days, the company received a highly credible report of a jailbreak from what David Sacks described as a "trusted partner." Rather than patch the vulnerability or de-deploy the model, Anthropic went to the government. The result: an export control order that effectively locks out every non-US person from some of the most capable AI systems ever built.
The US government didn't just restrict one company—it created a precedent that any advanced AI model can be treated as a controlled munition.
Amazon CEO Andy Jassy was among the tech leaders who raised concerns directly with Trump administration officials, pushing for the restrictions. Meanwhile, European political figures are calling this a "wake-up call" about depending on the US for AI technology.
The Sovereignty Problem
This isn't just about one company or one model. The US just demonstrated that it can—and will—control access to the most powerful AI systems through export control law. If your country doesn't have equivalent domestic AI capabilities, you're now dependent on US policy decisions for access to the bleeding edge.
The timing is notable: SpaceX just rented its Colossus 1 data center to Anthropic after internal teams struggled with Grok development due to latency issues. The same company building the most capable AI models is now explicitly prohibited from sharing them with roughly half the world's population.
The White House has indicated it won't extend these restrictions to other AI companies—at least for now. But the precedent is set. Future model releases will face this uncertainty: will the next Fable be restricted before it's even released?
What This Means
For AI developers outside the US, the writing is clear. If you need guaranteed access to the most capable models, you either need to build them yourself or accept that your access is subject to political winds. The $6.6B employee share sale that made OpenAI's Anthropic the biggest talent liquidity event in AI history now looks different when viewed alongside export controls that could make that talent less valuable to non-US organizations.
The question isn't whether AI will be restricted—it's whether the restriction is a bug or a feature. And whether the rest of the world has any say in the answer.