PANews reported on March 27th, citing Fortune, that Anthropic has acknowledged testing a new model, Claude Mythos, which represents a "leapfrog improvement" over its existing products and has already been made available to a small group of early adopters. Leaked internal drafts reveal that the model, codenamed "Capybara," outperforms the previous top-performing Claude Opus 4.6 across the board, showing significant improvements in software programming, academic reasoning, and cybersecurity tests. Anthropic has left nearly 3,000 unreleased assets in its unpublished data cache, including draft blog posts introducing Claude Mythos/Capybara and materials from a closed-door, invitation-only summit with European CEOs. The company claims the model "far surpasses other AIs" in cyberattack and defense capabilities and plans to initially deploy it to cybersecurity defenders, with limited public availability due to its high operating costs and risks.