SAN FRANCISCO :-
Anthropic shut down its two most powerful AI models for every customer Friday after the U.S. government ordered the company to block access for any foreign national, citing national security.
The order covers Claude Fable 5 and Claude Mythos 5, both released days earlier. Anthropic said in a statement it received the directive at 5:21 p.m. Eastern and disabled both models the same evening. A United States official confirmed the Commerce Department sent the letter, Bloomberg reported.
One detail sets this order apart from earlier export controls: it reaches inside Anthropic itself. The ban applies to any foreign national whether inside or outside the United States, including the company’s own non-citizen employees. With no way to screen those users out in real time, Anthropic said it had to switch the models off for everyone rather than attempt a partial block. Its other Claude models, including Claude Opus 4.8, are not affected, Fortune reported.
What the government ordered
The directive came from Commerce Secretary Howard Lutnick in a letter to Anthropic Chief Executive Dario Amodei, drafted with help from the Commerce Department’s Bureau of Industry and Security, according to an administration official cited by NBC News. It requires a license to export, re-export or domestically transfer the two models.
The Commerce Department did not respond to requests for comment from several outlets.
Anthropic said the letter did not spell out the specific concern. The company’s understanding is that officials believe someone found a way to bypass, or “jailbreak,” Fable 5’s safeguards in a way that could unlock the cybersecurity abilities of the underlying Mythos model. A jailbreak is a prompt or trick that gets a model to ignore its own safety rules. The worry, in this case, is that the trick could turn a consumer chatbot into a tool for finding software flaws that hackers could exploit.
An administration official told Axios the department acted after another company claimed it had jailbroken Mythos.
Anthropic’s pushback
Anthropic disputed the basis for the order. It said it reviewed a demonstration of the technique and found it surfaced only a small number of previously known, minor vulnerabilities. Other publicly available models, including OpenAI’s GPT-5.5, can find the same flaws without any bypass, the company said, and those rivals are not under the same export controls.
The company called the episode a likely misunderstanding. It said pulling a model used by hundreds of millions of people over one narrow flaw was the wrong response, and warned that the same standard, applied across the industry, would halt new model releases by every major developer, CNN reported.
Anthropic said it is complying while it works to restore access, and that it planned to share more within 24 hours.
Fable 5 vs Mythos 5
Fable 5 went on general release Tuesday, June 9. It was the first widely available model in what Anthropic calls its Mythos class, the tier the company describes as its most capable. Mythos 5, the less restricted version, had been limited to a small group of vetted partners under a security program Anthropic calls Project Glasswing. Both share the same underlying system and descend from an earlier model, Mythos Preview, announced in April.
| Claude Fable 5 | Claude Mythos 5 | |
|---|---|---|
| Availability | Public, general release | Limited to vetted partners |
| Program | Open to all customers | Project Glasswing |
| Safety filters | Blocks high-risk areas such as cybersecurity | Some limits removed |
| Launched | Tuesday, June 9, 2026 | Restricted, same period |
| Underlying model | Mythos (shared) | Mythos (shared) |
| Status after order | Disabled worldwide | Disabled worldwide |
A harder line in Washington
For years, U.S. export controls focused on the chips and equipment that power AI, not on the models themselves. Restricting foreign access to a model directly is a sharp escalation.
Earlier this month, President Donald Trump signed an executive order asking leading AI developers to hand over their most capable models for government cybersecurity testing before public release. The Fable 5 order is one of the first concrete moves under that line, and it will likely shape how other developers plan their launches.
The clash with Anthropic is not new. Relations broke down this year after the company refused to let its models be used for domestic surveillance and fully autonomous weapons. In March, the Pentagon labeled Anthropic a “supply chain risk,” a designation usually reserved for foreign adversaries, and barred defense contractors from using its Claude models in military work. Anthropic sued to reverse the label. That case is still going.
| When | What happened |
|---|---|
| Early 2026 | Anthropic refuses use of its models for surveillance and autonomous weapons; talks collapse |
| March | Pentagon labels Anthropic a “supply chain risk”; Anthropic sues to reverse it |
| April | Anthropic announces Mythos Preview |
| June 1 | Anthropic files confidential IPO paperwork |
| Early June | Trump signs order on pre-release model testing |
| June 9 | Fable 5 and Mythos 5 launch |
| June 12 | Commerce Department orders both models disabled for foreign nationals |
The IPO at stake
The order arrived late on a Friday, after U.S. markets closed, and three days after the two models launched. The timing is awkward for Anthropic.
The company filed confidential IPO paperwork with the Securities and Exchange Commission on June 1, CNBC reported. A confidential draft discloses no financials publicly, but figures from Anthropic’s late-May funding round are known: a $47 billion annualized revenue run rate and a $965 billion valuation after a $65 billion Series H, per Fortune. That valuation briefly made it the most valuable private AI company, ahead of OpenAI.
A government move that singles out Anthropic’s flagship models could make some investors question whether it can stay at the front of AI development.
There is a competitive cost too. U.S. buyers are already drifting toward open-weight models, many of them Chinese. A March report from the U.S.-China Economic and Security Review Commission found that 80 percent of U.S. startups were using Chinese open-source models, and that Chinese labs’ share of model downloads on Hugging Face rose from about 1.2 percent at the end of 2024 to roughly 30 percent a year later. None of those open-weight models restricts who can download or fine-tune them.
How dangerous is Mythos, really?
The stakes are not only commercial. Mythos is reportedly already in government hands. The Financial Times reported that Anthropic embedded about half a dozen engineers at the National Security Agency to help adapt the model for offensive cyber operations, work earlier flagged by Axios. FT’s sources said it was unclear whether the engineers or the model were being used in live operations, as opposed to setup and customization.
That arrangement is itself contested, given the Pentagon’s separate ban on Anthropic as a vendor.
Some researchers also question the threat. Independent testers have found that cheaper open-source models can match much of Mythos’s ability to find software vulnerabilities, and critics say a closer look at Anthropic’s own figures shows fewer serious exploits than its marketing implies. That has fed a view among some observers that the national security framing serves Anthropic’s image as much as it reflects any real risk.
For now, the company’s most advanced models are offline for everyone. Anthropic says that is temporary. How long it lasts may depend less on the strength of the alleged jailbreak than on how fast Anthropic and Washington can repair a relationship that has frayed all year.
Frequently asked questions
What happened to Anthropic’s Fable 5 and Mythos 5? Anthropic disabled both models worldwide Friday, June 12, after the Commerce Department barred access by any foreign national. Because it could not filter foreign users in real time, it switched the models off for all customers.
Why did the government order the shutdown? It cited national security. Anthropic says officials believe someone found a jailbreak that could bypass Fable 5’s safeguards and unlock the cybersecurity abilities of the underlying Mythos model. Axios reported the action followed another company’s claim that it had jailbroken Mythos.
Could this affect Anthropic’s IPO? Possibly. Anthropic filed confidential IPO paperwork June 1. Reported figures include a $47 billion run rate and a $965 billion valuation. A government action against its top models could raise investor questions about its lead.
What is the difference between Fable 5 and Mythos 5? Both run on the same system. Fable 5 is public and includes filters that block high-risk responses such as cybersecurity. Mythos 5 is less restricted and went only to vetted partners under Project Glasswing.
Are other Claude models affected? No. The order applies only to Fable 5 and Mythos 5. Other Claude models, including Claude Opus 4.8, stayed online.
When will the models come back? Anthropic has not given a date. It said it is complying while working to restore access, and planned to release more within 24 hours.
Leave a Reply