SAN FRANCISCO ? Anthropic has begun a limited preview of its new Mythos model for cybersecurity partners, formally introducing a system that had already drawn attention after an internal leak described it as one of the company's most powerful AI models.

In TechCrunch's April 7 report, Anthropic said Mythos would be tested by 12 partner organizations under a new Project Glasswing initiative focused on defensive security work and the protection of critical software.

The preview followed an earlier internal leak. TechCrunch wrote that Anthropic had accidentally exposed thousands of internal files, including material that referred to Mythos under an earlier codename and described it as significantly more capable than the company's public models.

Anthropic's subsequent attempt to clean up related code exposure also created fresh controversy. In a separate report, TechCrunch said the company temporarily helped trigger takedowns affecting thousands of GitHub repositories while trying to remove leaked material.

The official preview therefore landed in a very different context from a typical product launch, with Anthropic presenting Mythos as a security tool even as it worked to contain the fallout from recent disclosure failures.