Anthropic has disabled two of its newest and most powerful AI models worldwide, just four days after launching them, following an order from the United States government.
The company said on Friday that the U.S. government had sent a letter directing it to suspend all access to Claude Fable 5 and Claude Mythos 5 for any foreign national, whether they are inside or outside the United States. The order also covers Anthropic employees who are not U.S. citizens.
What The Order Covers
Anthropic had already treated this model with caution. When it first introduced Claude Mythos in April 2026, the company said the model could autonomously find and chain together zero-day exploits across major operating systems and browsers, and kept access limited to around 40 vetted partners through a program called Project Glasswing rather than release it widely.
But that quickly changed on June 9, when Anthropic opened a version of the same system, Claude Fable 5, to the public with added safeguards. Claude Mythos 5, the version with those cybersecurity safeguards removed, stayed restricted to Project Glasswing partners. The U.S. government’s order covers both versions.
However, all of Anthropic’s other models, including Opus, Sonnet and Haiku, remain available and are not affected.
The Security Concern Behind The Decision
Anthropic said the government’s directive cited national security concerns tied to a jailbreaking method that could unlock Mythos 5’s cybersecurity capabilities through Fable 5. Those capabilities include finding and chaining together zero day software vulnerabilities, flaws that have not yet been publicly disclosed or fixed.
The company described the jailbreak in question as narrow, meaning it works in one specific case rather than defeating the model’s safeguards entirely. Anthropic also said it believes a similar technique could likely work on other publicly available models that face no such restrictions.
Anthropic’s Response
Anthropic said it complied with the directive immediately but disagrees with it. “We believe this is a misunderstanding,” the company said in its statement, adding that it is working to restore access for foreign users as soon as possible.
This is not the first clash between Anthropic and the current administration. Earlier this year, the Pentagon labelled the company a supply chain risk during a separate dispute, a decision that blocked Anthropic from government contracts.
What This Means For Global AI Regulation
The order marks one of the first times a government has used export control powers to restrict an AI model itself, rather than the chips or computing infrastructure behind it. Earlier export controls in the AI sector focused mainly on hardware.
For now, Fable 5 and Mythos 5 remain offline everywhere while Anthropic works through the matter with the U.S. government. This episode points to a deeper tension shaping the AI industry. Companies are under pressure to ship frontier models quickly and win the AI arms race, especially as rivals push out their own advances, yet governments are growing more willing to use export controls to restrict access to that same technology in the name of national security. How AI labs navigate between racing ahead and answering to national governments could decide what gets released next and who gets to use it.
