According to tech media Testingcatalog, Anthropic appears to be preparing to release a preview of its latest model, Claude Mythos. This news has attracted widespread attention, as Mythos is considered one of the company's largest AI models to date.

According to sources, users previously briefly found an option called "claude-mythos-1-preview" in the public versions of Claude and Claude Security, but the option was later removed. This suggests that Anthropic may have entered the pre-launch testing phase of the model. The development of Claude Mythos dates back to April 7th, with the official positioning it as a cutting-edge model for computer security tasks. Compared to the existing flagship model Opus 4, Mythos has made significant improvements in code understanding and autonomy.

However, despite its strong capabilities, Anthropic has not rushed to fully open up Mythos. The company previously stated that this model is capable of automatically acquiring professional-level cyber attack methods, and releasing it could pose serious risks to digital infrastructure. This potential threat has caused many people to worry about the public release of Mythos.

Notably, some users briefly saw an option to enable Mythos in the public version of Code, and a model with the same name also appeared in Claude Security. However, as of now, Anthropic has not officially announced a specific release date.

Meanwhile, Anthropic confirmed that it is advancing a new project called "Glass," which aims to collaborate with other companies to protect global critical software systems from AI-driven vulnerabilities. This initiative has not used the Claude Mythos preview version and has already partnered with up to several organizations. Anthropic demonstrated an open-source vulnerability dashboard to summarize different severity level vulnerabilities found in the Mythos preview version.