Anthropic Blocks Foreign Access to Claude 5 Under US Export Rules

Published by James Harris on June 13, 2026June 13, 2026

The Trump administration has ordered Anthropic to cut off foreign nationals from its two newest AI models, Claude Fable 5 and Mythos 5, citing national security concerns. Anthropic complied immediately, suspending access for all customers while it works through the directive, including locking out its own foreign-born employees who helped build the models.

The comparison that keeps surfacing among critics is the 1990s encryption wars, when the US government classified cryptographic software as a munition under arms export law. Those controls collapsed within a decade, partly because the math was already public and partly because enforcement was practically absurd. The parallel here is sharper than it sounds: Anthropic is now defending its safety record to regulators while simultaneously being told which humans are permitted to access what it built. The “deemed export” rule, which treats a foreign-born employee viewing a model as an export of that model to their home country, was designed for physical hardware. Applying it to weights and inference APIs in 2025 stretches the legal framework past its original intent. Dean W. Ball’s observation that the administration has approved advanced chip sales to China while throttling allied nations’ access to American AI models is the clearest articulation of why critics are calling the policy incoherent rather than merely aggressive.

Anthropic launched Fable 5 as its general-access flagship and Mythos 5 as a restricted variant on June 9. The export restriction landed three days later.

That timing matters for reasons beyond the obvious. Anthropic is reportedly targeting a valuation near $1 trillion in an anticipated IPO, and the episode has introduced a category of regulatory risk that prospective public-market investors had not yet priced into frontier AI companies: the possibility that the US government can effectively nationalize access to a private company’s core product on short notice. Internationally, the reaction has moved past frustration into something closer to policy urgency. Vasant Shetty’s observation that India, the second-largest market for both ChatGPT and Anthropic’s products, is now “at the mercy of a foreign government” reflects a concern that was theoretical six months ago and is now operational. Advocates for open-source and locally hosted models are already using this moment as a proof point, and that argument is harder to dismiss than it was before June 12.

Dario Amodei addressed the gap between AI development speed and policy infrastructure in an essay published on June 10, one day before the restriction landed. Whether that was foresight or coincidence, the sequence makes it harder for Anthropic to argue it was caught entirely off guard, and harder for the administration to argue the timing was unrelated to the company’s public posture on autonomous weapons and surveillance earlier this year.