In a dramatic turnaround, Anthropic says the US government has rolled back weeks-long export restrictions that had blocked the AI startup from exporting its über-powerful Claude Mythos 5 and Fable 5 models over national security concerns.

“We’ve received notice that the Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5," Anthropic posted on X late Tuesday, adding that it would begin restoring access on Wednesday.

The company also wrote that it was grateful for the patience of its Claude users “and to everyone who worked with us on redeploying the models.”

Anthropic said it was forced to "abruptly disable" the vulnerability-hunting Mythos 5 and its watered-down public-facing counterpart, Fable 5, after receiving the export-control order on June 12th.

The move comes less than three weeks after the US Commerce Department ordered the company to suspend access for foreign nationals, citing concerns the advanced AI models could fall into the wrong hands and threaten national security.

Last Friday, Washington partially reversed the order, allowing Anthropic to release its upgraded Mythos 5 model to several "trusted" US organizations.

Keven Knight, CEO at Talion Cyber Security, said given the partial easement last week, it was only a matter of time before the Commerce Department reversed its order for Fable as well.

“Restricting access to the platforms would only leave Western organizations on the back foot – and at a time when the AI arms race is really heating up, this would be dangerous,” he explained.

Knight also noted that, although it's unclear how Anthropic appeased the US government and addressed its security concerns, it appears there is now “collaboration between the two and an agreement to work together more closely on future updates.”

Cybernews previously reported that although Fable 5 and Mythos 5 share the same underlying frontier AI model, Fable 5 was built with extra safety restrictions that can block or redirect high-risk cyber requests to the older Claude Opus 4.8 model.

Why Washington fears Mythos

Considered one of the most powerful bug-hunting AI models currently in existence, Mythos has proven capable of finding software flaws at lightning speed.

This has raised major concerns among security experts, the Trump administration, and Anthropic itself that the cybersecurity model could be used by foreign US adversaries, including China and Russia, to rapidly identify and exploit countless unknown vulnerabilities before defenders can identify and patch them.

In fact, just last week, it was revealed that during testing by US intelligence agencies, the frontier model essentially “broke into almost all of our classified systems, not in weeks, but in hours," according to testimony from Senator Mark Warner before Congress earlier this month.

The revelation has prompted US lawmakers to propose new legislation that, if passed, would establish AI-related incident reporting rules for the first time.

Knight believes last week's unveiling by China's Beijing-based 360 Security Technology of its own powerful Mythos rival likely contributed to the decision.

“The US could not afford to limit access to Anthropic's platforms for much longer,” he tells Cybernews.

However, Knight warns that now organizations have been given the green light to reinstate access to Fable and benefit from its security capabilities, enterprises “must ensure the platform is governed properly, and understand the reasoning behind its actions and conclusions.”

Access to Mythos and Fable was officially banned across all US federal agencies almost as soon as they were released.

Project Glasswing – which has now ballooned to more than 200 technology and infrastructure organizations, including Amazon, Google, Microsoft, Apple, and Cisco – allows those companies permission to test Mythos’ extraordinary vulnerability-detection mechanisms and autonomous security patching against their own systems.

The export reversal also comes just days after Anthropic accused Chinese tech giant Alibaba of attempting to syphon Mythos’ capabilities in what it described as “the largest known AI model distillation campaign.”

Mythos Preview entered limited testing with approved organizations in late February, Anthropic publicly unveiled it under Project Glasswing on April 7th before launching Mythos 5 and then its guardrailed version, Fable 5, in June.

Unlock more exclusive Cybernews content on YouTube