Anthropic to Restore Fable 5 and Mythos 5 After US Lift

Anthropic will restore public access to its Claude Fable 5 and Mythos 5 models after the US government removed export controls that had blocked access since June 12.

The export controls followed a report in which researchers demonstrated a method to bypass Fable 5’s safeguards and prompt the model to identify software vulnerabilities. Anthropic pulled access to the models on June 12 while officials reviewed the findings.

Anthropic plans to redeploy the models with updated safety classifiers designed to block more cybersecurity-related tasks. The company wrote, “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.”

Howard Lutnick posted on X, “Over the past two weeks, we have worked closely with Anthropic to analyze and approve Fable 5 to ensure alignment across the US Government and to strengthen America’s leadership in AI.” White House Chief of Staff Susie Wiles wrote on X that the government’s priority is to “get the best [AI] tech deployed as quickly and safely as possible.”

Anthropic said it will expand cooperation with US agencies and with partners in Project Glasswing, a coalition that includes Amazon, Microsoft and Google, to develop a framework for assessing the severity of AI jailbreaks. The company added the work will include pre-release access to models and safeguards for evaluation, information sharing on jailbreaks and misuse, and dedicated resources for joint research.

The report that prompted the controls included work by Amazon researchers who showed a way to make Fable 5 reveal software flaws. An independent researcher also published screenshots showing a bypass of the model’s guardrails within 48 hours of the model’s launch.

Anthropic noted that weaker models can produce similar outputs under some conditions. The company said the revised classifiers will be paired with increased government testing and oversight before future releases and that it will return the models to public availability after redeployment with the updated safety systems.