Related: Claude Fable 5 and Mythos 5 Remain Offline Under a US Export Order
After weeks of negotiations, the White House has permitted Anthropic to restore access to its most advanced model, though only for a carefully vetted group of US companies and government agencies rather than the public at large. The decision, reported Friday, partly resolves a standoff that had pulled one of the most capable AI systems in the world offline for everyone, and it marks the moment the government moved from blocking the model to deciding exactly who is allowed to use it. Anthropic can now release its Mythos 5 model to roughly 100 institutions, according to the reporting, while talks over its companion model, Fable 5, reportedly continue.
The reversal closes the most acute phase of a fight that began about two weeks earlier. On June 12, the Commerce Department’s Bureau of Industry and Security, under Secretary Howard Lutnick, issued an export-control directive citing national security authorities and ordered Anthropic to suspend all access to Fable 5 and Mythos 5 for any foreign national, a requirement so broad that complying meant switching the models off for every customer worldwide. The order arrived in the evening and the models were gone by midnight, an abrupt outage for systems Anthropic had launched only days before as its most powerful to date.
At the center of the dispute was a security concern rather than the models’ everyday abilities. The government had become aware of a method for bypassing a safeguard meant to stop Fable 5 from being used to identify software vulnerabilities, the kind of weakness that, in the wrong hands, could be turned into a cyber weapon. Anthropic disagreed sharply with how the government weighed that risk, describing the action as resting on a misunderstanding and arguing that a single narrow jailbreak should not be grounds for recalling a model already in commercial use. The company said throughout that it was working to restore access as soon as possible.
The capability that worried officials is real and unusually concrete. Mythos had been released under a controlled-access program Anthropic calls Project Glasswing, offered to a select group of technology and financial organizations rather than the open market, precisely because of how powerful it is at security work. In internal testing the model identified thousands of previously unknown zero-day vulnerabilities across every major operating system and browser, a demonstration of exactly the dual-use problem at the heart of the standoff. A tool that can find flaws that fast is invaluable to defenders patching their systems and dangerous to anyone who would exploit them instead.
That tension explains why the restoration came with conditions rather than a simple reversal. The White House had been working to put safeguards in place before widening access, coordinating with model providers, other industry partners, and the intelligence community to make sure the right guardrails existed, and reporting suggests agencies may receive a modified version of the model rather than the original. The result is a release defined by control. Mythos is not coming back for the public, it is being switched on for a vetted list of around a hundred companies and federal agencies that the government is comfortable with, a very different thing from a model freely available to anyone who pays.
The episode lands at the intersection of two forces that have shaped the year, the rapid advance of frontier AI and the government’s growing willingness to treat that capability as a matter of national security. It is not an isolated case. Only a day earlier it emerged that the administration had asked OpenAI to limit the release of its own new model, GPT 5.6, to government-approved customers, a near mirror of the access controls now governing Mythos. Together the two cases suggest a pattern is forming in which the most powerful models reach the market through a gate the government holds rather than a public launch the company controls.
For Anthropic, the partial restoration is a relief and a precedent it may not welcome. Its most advanced system is usable again by the customers that matter most to it, but the path back ran through a negotiation with Washington over who may touch the model, and Fable 5 remains offline pending further talks. The broader signal for every frontier developer is hard to miss. A model can be switched off by order and switched back on only on terms the government sets, and the companies building these systems no longer have the final say over whether their most capable products stay available. How the remaining negotiations over Fable 5 land, and whether a modified Mythos satisfies both sides, will shape how that new balance of power between the AI labs and the state settles from here.
Related on Entrelligence: the US lifting export controls on Claude Fable 5, and a Chinese open model matching Anthropic’s Mythos.
Your First 10 AI Skills
10 practical AI skills, copy-paste prompts and a 7-day plan to start using AI with confidence.
Download the guide →