Anthropic, the AI safety and research company behind the Claude family of models, reached an agreement with the Commerce Department to restore public access to its Fable 5 and Mythos 5 models, deploying them with new guardrails and classifiers designed to block jailbreak.

 

What happened

Anthropic announced on Tuesday that the Commerce Department lifted export controls that had blocked sales of Fable 5 and Mythos 5 to foreign companies and individuals. Fable 5 and Mythos 5 are two of Anthropic's newest and most powerful AI models, ranked above the company's Claude Opus models. They are built on the same technology, the difference is that Fable 5 has extra safety restrictions around biology and cybersecurity.

The company also restored U.S. user access to the models. Anthropic reached the agreement after weeks of negotiation with the White House and Commerce Department. Secretary of Commerce Howard Lutnick confirmed the decision on X. Anthropic paired the restored access with newly trained safety classifiers built to target and block the behavior described in a threat intelligence report that had triggered the original restrictions.

 

The backstory

The Trump administration imposed the export controls after Amazon published a threat intelligence report claiming to have jailbroken Fable's cybersecurity capabilities. Officials grew concerned that releasing Fable 5 would let users unlock cybersecurity and other capabilities that Anthropic warned could cause serious harm on the open internet if misused. The Amazon report convinced administration officials that such jailbreaks were imminent, prompting Commerce to act just days after Fable's release, despite a White House executive order from the prior month that had established a voluntary national security testing arrangement with Anthropic and OpenAI.

 

Going deeper

Anthropic pushed back on the severity of the Amazon report's findings. The company tested equivalent and lesser models, including ChatGPT 5.5, Claude Opus 4.8, and Kimi K2.7, and found they could identify the same vulnerabilities Fable identified in the Amazon report. Anthropic also found roughly six existing models capable of producing the same proof-of-concept code as Fable. Anthropic maintains it has not seen a jailbreak affecting the model's restrictions on cybersecurity and biology work, though it described the Amazon incident as a "borderline case." Some cybersecurity professionals separately complained that Fable 5's existing safeguards were blocking routine defensive cybersecurity work along with malicious use cases. The federal Center for AI Standards and Innovation stress-tested Anthropic's new classifiers before their deployment.

 

What was said

Secretary Lutnick wrote on X that Commerce and Anthropic had "worked closely" over two weeks to "analyze and approve Fable 5."

Anthropic's blog stated that the Amazon-reported technique "did not expose any unique Mythos-level cyber capabilities," and characterized it as a case where safeguards block some low-risk tasks "out of an abundance of caution."

Christopher Padilla, a former Assistant Secretary of Commerce for export administration under George W. Bush, called the resolution "good news" but said the episode shows "the risks of ad hoc, transactional policymaking." In a LinkedIn post, Padilla also said the same Bureau of Industry and Security office that restricted Fable and Mythos maintains a "permissive policy for exporting high-end AI semiconductors to China - in exchange for a cut of the take."

 

Why it matters

This series of events shows how a third-party threat report can trigger a federal export freeze on a commercial AI model within days, even when the capability isn't new. Anthropic's own testing found that several other publicly available models, including competitors' products, could reproduce the same results that alarmed regulators, meaning the restriction targeted one company's model without addressing the capability across the industry. For companies building on frontier AI models, the case is a reminder that regulation can change quickly, even under a recently signed national security testing framework.

 

The bottom line

Fable 5 and Mythos 5 are back online for U.S. and international users, now paired with classifiers meant to close the specific gap the Amazon report surfaced.

Read also: Is Claude Fable HIPAA compliant?

 

FAQs

What is Mythos 5, and how does it differ from Fable 5?

They share the same underlying model, but Fable 5 includes additional safety measures around biology, cybersecurity, and LLM research and development.

 

What is the Center for AI Standards and Innovation (CAISI)?

CAISI is the federal body within the Commerce Department focused on evaluating and setting standards for AI systems, including national security-related testing.

 

What is the Bureau of Industry and Security (BIS)?

BIS is the Commerce Department office responsible for administering U.S. export controls, including restrictions on advanced technology like AI models and semiconductors.

 

What are export controls?

Export controls are government restrictions that limit which products, technologies, or services can be sold or transferred to foreign entities, often for national security reasons.