Anthropic Restores Access to Fable 5 After U.S. Restrictions Lifted

Anthropic has announced the restoration of access to its Claude Fable 5 and Claude Mythos 5 AI models after the U.S. Department of Commerce lifted export restrictions on June 30. Access to Fable 5 will be reinstated globally starting July 1, with utilization available on Anthropic’s platforms and through partner cloud services like Amazon Web Services and Microsoft Foundry expected soon.

“We appreciate our users’ patience and the collaborative effort with the government,” said Anthropic.

In a focus on cybersecurity, Anthropic revealed it will deploy a new classifier for the Fable 5 model to enhance its ability to identify and block potentially harmful tasks. The firm noted that this adjustment follows productive discussions with U.S. officials, stressing the need for improved security measures in AI applications. Notifications will alert users if any requests are blocked, redirecting them to Claude Opus 4.8 instead.

While access to Mythos 5 has begun for select U.S. organizations, Anthropic plans to maintain ongoing collaboration with regulatory bodies to ensure the safety and efficacy of its tools. This includes sharing insights on security breaches, developing protective measures, and conducting joint research initiatives.

“It is likely impossible to make any AI model completely resistant to jailbreaks,” stated Anthropic, acknowledging challenges in ensuring absolute security.

As a proactive measure, Anthropic is also collaborating with other tech giants under Project Glasswing to create a standardized framework for assessing the risks associated with AI jailbreaks. This framework evaluates threats based on their potential to enhance an attacker’s capabilities, the scope of malicious tasks affected, ease of execution, and replicability of the jailbreak methods.

For severe breaches, Anthropic commits to prompt protective action upon threat confirmation, while also establishing a monitoring team focused on tracking jailbreak dissemination.

In related developments, OpenAI launched its own model, GPT-5.5-Cyber, for vulnerability management during the same period when Anthropic faced restrictions.