Anthropic restores Fable 5 after export controls lifted

• Anthropic will restore public access to its Claude Fable 5 and Mythos 5 AI models after the US government lifted temporary export controls.
• The company has added new security classifiers designed to block a broader range of cybersecurity-related requests following concerns over the models' safeguards.
• Anthropic is also expanding cooperation with the US government and technology partners to strengthen AI safety testing and jailbreak standards.

Anthropic will restore public access to its Claude Fable 5 and Mythos 5 artificial intelligence models after the US government lifted export controls that had restricted access since June 12 following security concerns.

The restrictions were introduced after researchers demonstrated a method of bypassing Fable 5's safeguards, prompting Anthropic to suspend access while the company and US officials reviewed the model's security protections.

“After a series of productive conversations with the US government, we’re redeploying the model with a new set of classifiers to target and block more cybersecurity tasks,” Anthropic said.

The updated models will include additional classifiers designed to prevent a wider range of cybersecurity-related requests, while Anthropic said it will expand information sharing, model testing and safeguard evaluations with the US government before future releases.

US Secretary of Commerce Howard Lutnick said the government worked with Anthropic to review Fable 5, while White House Chief of Staff Susie Wiles said the administration's priority is to deploy advanced AI technology as quickly and safely as possible.

Anthropic also said it is developing a framework with Amazon, Microsoft, Google and other partners through Project Glasswing to establish common standards for assessing the severity of AI jailbreaks and improving cybersecurity protections.

The temporary restrictions followed reports that researchers had bypassed Fable 5's safety guardrails, highlighting the growing focus by governments and AI developers on balancing public access to advanced models with national cybersecurity concerns.