US government export control directive suspending foreign national access to Anthropic Fable 5 and Mythos 5, including Anthropic employees
The US Government Just Cut Anthropic Employees Off From Their Own Models
Read that sentence a few times. Let it land.
The US government has issued an export control directive suspending all access to Anthropic’s Fable 5 and Mythos 5 by any foreign national, whether they’re inside the United States or not. That includes foreign national employees at Anthropic itself. People who helped train these models, ran the evals, wrote the safety research. Cut off.
This is a big deal. I don’t think the industry has fully absorbed what just happened.
⚙️ What Actually Occurred
Anthropic confirmed the directive directly via their official account: “The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.”
The models themselves are remarkable. Andrej Karpathy called Fable 5 a “major-version-bump-deserving step change forward,” noting it’s SOTA on everything by a margin, and emphasized the qualitative difference goes beyond benchmarks. These are not incremental models. They’re the ones the government just decided need to be treated like sensitive military hardware.
The Borderless Assumption Is Dead
For years, the working theory in AI was that progress couldn’t be contained. The talent is international. Research papers go on arXiv the day of publication. Benchmarks don’t ask for citizenship. The implicit social contract between AI labs and the global research community was openness by default.
That assumption is gone. Export controls on frontier AI models mean the US government now views certain model weights and capabilities the same way it views encryption algorithms, advanced semiconductors, and defense technology. That’s not a subtle shift. That’s a category change.
The Human Cost Nobody Is Talking About
I keep coming back to this specific detail: Anthropic’s own foreign national employees are affected. These aren’t adversaries. They’re researchers who chose to work at an American AI safety company. Some of them probably took pay cuts to be there because they believed in the mission.
Now they can’t access the models they spent months building. That’s genuinely disorienting, and I think the practical and morale consequences inside Anthropic are going to be significant and underreported.
🔍 What This Signals About Government Intent
Anthropic has been positioning itself as the policy-friendly AI lab. Dario Amodei published an essay on closing the gap between AI progress and policymaking institutions. Anthropic launched a $150 million national fellowship program. They published an Advanced AI Framework explicitly stating governments should have authority to block unsafe model releases. They contributed $200 million to fund evaluations of AI economic policy ideas.
They were building the relationship with Washington. And Washington responded by issuing a directive that cuts through their own workforce.
That’s not a rebuke. It’s actually the logical extension of everything Anthropic asked for. If you argue governments need real authority over frontier AI, don’t be surprised when they use it. Anthropic wrote the framework. The government is using it.
Where This Goes
The talent implications are real and probably irreversible in the short term. If you’re a brilliant AI researcher with a non-US passport, you now have a concrete, documented reason to think twice before joining a US frontier lab. Some of that talent goes to European labs. Some goes to Chinese labs. The export control intended to protect US AI advantage may accelerate the redistribution of the people who create that advantage.
I don’t have a clean answer for how you square that circle. But I’d note that the government imposed this directive on the most safety-conscious major AI lab, the one most aligned with federal oversight, before figuring out that problem.
That ordering matters.
Sources:
#AI #ArtificialIntelligence #ExportControls #Anthropic #NationalSecurity #AIPolicy #MachineLearning
