Anthropic will make a locked-down version of Mythos available for cybersecurity tasks, months after warning that the powerful artificial intelligence model could identify and exploit vulnerabilities in critical software.
The new model, called Fable 5, will begin to be distributed this Tuesday (9) with protections that prevent responses to certain types of requests, including topics related to cybersecurity and biology. In these cases, Anthropic said its Claude chatbot will forward responses to another model, called Opus 4.8.
Anthropic is also releasing the same model, but without some of these safeguards, as a new version of Mythos called Mythos 5. It will be available to groups authorized to use the version with cyber capabilities through an initiative called Project Glasswing. Last week, Anthropic added another 150 organizations to the list of companies with access to Mythos, bringing the total to about 200.
Study abroad
Upgrade your career!
Mythos has become a key focus for the San Francisco-based company in recent months as it moves toward an initial public offering (IPO). The company took the unusual step of restricting access to the model to select partners, citing concerns that it would be able to identify and exploit vulnerabilities “in all major operating systems and all major web browsers when directed by a user to do so.” At the same time, like its rivals, Anthropic is also working on developing more capable models for a range of profitable tasks, such as programming, finance and cybersecurity.
“We wanted to ensure that in non-cyber use cases, the priority was to launch Fable safely as quickly as possible,” said Dianne Penn, head of project management for Anthropic’s research and labs. “So we’re releasing this part of Fable first, while we continue to work on the broader cybersecurity use cases.”
According to the company, Fable 5 is designed to be better at programming and other professional tasks, especially solving complex problems over longer periods of time than previous models.
Continues after advertising
In a test of the new software, payments processor Stripe was able to complete an extensive software engineering task in one day that would have taken a team two months to complete manually, Anthropic said. The company also said that a hypothesis generated by Mythos about a new mechanism of a bacterial protein E. coli was confirmed in a scientific article from a laboratory that studied the same topic.
To test whether Fable 5’s protections would hold up, Anthropic said it internally conducted an bug bounty to find ways to so-called jailbreak the model, that is, bypass its security barriers. In more than a thousand hours of testing, red team experts have not found any universal method to game the system.
Penn said Anthropic will continue to expand the number of groups with access to the cyber-capable version of Mythos through Project Glasswing and will look for “other ways to allow trusted partners access for cybersecurity purposes.”
© 2026 Bloomberg L.P.
