US lifts curbs on Fable 5, Mythos 5, Anthropic restores access to AI models

Anthropic has restored access to Fable 5 and Mythos 5 after the US eased export controls, introducing stronger safeguards against AI jailbreaks

Artificial intelligence major Anthropic has restored access to its advanced AI models, Fable 5 and Mythos 5, after the US government lifted export controls that abruptly halted access earlier this month over national security concerns.

Fable 5 will become available to users globally starting July 1 across Anthropic's platforms, including Claude.ai and Claude Cowork, the company said in a blog post.

On the other hand, access to Mythos 5 resumed for select US organisations following government approval on June 26.

Anthropic said it is now coordinating with the US government to expand access to broader domestic and international partners under its Project Glasswing initiative.

 

"After a series of productive conversations with the US government, we're redeploying the model (Fable 5) with a new set of classifiers to target and block more cybersecurity tasks.

"We've also begun drafting a consensus framework-with Amazon, Microsoft, Google, and other Glasswing partners-for assessing the severity of AI jailbreaks and how AI developers should respond to them," Anthropic said in a post on X.

India is Anthropic's second-largest market. The AI firm recently signed a deal with TCS to equip 50,000 employees with its models. It also announced a collaboration with Infosys to deploy advanced enterprise AI solutions using the Claude family of models.

Furthermore, select Indian government agencies and private firms were recently granted access to the Mythos model under a cybersecurity initiative called Project Glasswing, according to sources familiar with the matter.

The initial US directive followed the discovery of a jailbreak by Amazon researchers. This method bypassed Fable 5's safeguards, prompting the model to identify software vulnerabilities and demonstrate how to exploit them.

To resolve the issue, Anthropic worked with the US government to train an "improved safety classifier" that targets and blocks the specific bypass technique in over 99 per cent of cases.

Anthropic said it is scaling up its collaboration with the US government on model testing and safeguards. This will include pre-release access to models and safeguards for evaluation, information sharing on jailbreaks and misuse and dedicated resources for joint research.