Anthropic PBC is widely releasing a version of Mythos that will be blocked from carrying out cybersecurity tasks, months after warning that the powerful artificial intelligence model could spot and exploit vulnerabilities in critical software. 

The new model, called Fable 5, is set to be rolled out on Tuesday with guardrails that prevent it from responding to certain types of queries, including those related to cybersecurity and biology. 

In those cases, Anthropic said its Claude chatbot will route responses through a different model called Opus 4.8. Anthropic is also releasing the same model, without some of the safeguards, as a new version of Mythos called Mythos 5.

It will be available to the groups that can use the cyber-capable model through an initiative called Project Glasswing.

Last week, Anthropic added 150 additional organizations to the list of firms with access to Mythos, bringing the total to about 200. 

Mythos has emerged as a focal point for the San Francisco-based company in recent months as it pushes toward an initial public offering.

The company made the unusual decision to restrict access to the model to select partners, citing concerns that it can identify and exploit vulnerabilities “in every major operating system and every major web browser when directed by a user to do so.”

Like its rivals, however, Anthropic is also working to develop models that are more capable at a range of lucrative tasks, including coding, finance and cybersecurity. 

“We wanted to make sure for non-cyber use cases, we really prioritized safely releasing Fable as soon as possible,” said Dianne Penn, head of project management for Anthropic’s research and labs. 

“That’s why we we’re bringing this part of the Fable piece first while we continue to work on the general cyber use cases.”

Fable 5 is designed to be better at coding and other professional tasks, particularly solving tricky problems over a longer period of time than prior models, the company said. 

In a test of the new software, payment processor Stripe was able to complete a lengthy software-engineering task in a day that would have taken a team two months to do manually, Anthropic said.

The company also said a hypothesis that Mythos generated regarding a new mechanism for an E. coli protein was confirmed in a research paper by a lab that was studying the same issue.

To test whether Fable 5’s guardrails held up, Anthropic said it ran an external bug bounty program to find ways to jailbreak the model, or get around its safeguards. In over 1,000 hours of testing, red teamers didn’t find any universal jailbreaks. 

Penn said Anthropic will continue to expand the number of groups that have access to the cyber-capable version of Mythos through Project Glasswing and take “other approaches to be able to have trusted partners access for cybersecurity.”

More stories like this are available on bloomberg.com

Published on June 10, 2026