Fortune reports Anthropic left details of its unreleased AI model “Mythos” exposed in a public data store on March 26, 2026. Cybersecurity researchers Roy Paz from LayerX Security and Alexandre Pauwels from the University of Cambridge found the info, including a draft blog post, in an unsecured cache linked to Anthropic’s content management system.
What got leaked about Mythos
The draft post called Mythos (also referred to as Capybara) the company’s most powerful model yet—a “step change” over previous ones like Claude Opus 4.6. It scored much higher on software coding, academic reasoning, and cybersecurity tests. Anthropic described it as trained and ready for careful rollout due to its strength, especially in cyber tasks that could help defenders but also aid attackers.
Fortune reviewed the materials and noted plans for early access with select customers first, given the model’s high costs and risks. The leak included around 3,000 unpublished assets tied to Anthropic’s blog.
Anthropic’s response
After Fortune alerted them, Anthropic locked down the data store. A spokesperson blamed “human error” in CMS setup and confirmed they are testing the model with early customers. “We’re developing a general purpose model with meaningful advances in reasoning, coding, and cybersecurity,” they said, calling it their most capable to date. No core systems or customer data was affected, they added. Fortune’s follow-up covers the CMS glitch.
This marks the first public hint at Mythos, though via mishap, not announcement. CNBC reports cybersecurity stocks fell following news of the powerful new model.
