Anthropic is testing a new artificial intelligence model, codenamed “Claude Mythos,” which the company describes as a “step change” in performance compared with its previous releases, according to a report by Fortune.
The existence of the model became public after a configuration error exposed roughly 3,000 unpublished files, such as draft blog posts, images, PDFs, and internal documents, on a publicly accessible data cache linked to Anthropic’s content‑management system. Among the files was a draft post detailing Mythos, including its capabilities in reasoning, coding, and cybersecurity.
Anthropic told Fortune that Mythos is a general‑purpose AI system expected to surpass prior models in performance. Internal documents reportedly suggest the model could achieve notably higher scores in coding, academic reasoning, and cybersecurity tasks, making it potentially the company’s most capable AI to date.
The leak also raised concerns about security risks. Anthropic has acknowledged that Mythos’s advanced capabilities could lower the barrier for sophisticated cyberattacks if misused, per Fortune’s report.
The company said the exposure was caused by human error in the CMS configuration, that the content was promptly removed, and that no core infrastructure or customer data were affected.
Rather than a full public launch, Anthropic is currently testing Mythos with a limited group of early‑access customers. The company says this cautious rollout allows it to refine safety measures and manage deployment in enterprise and government contexts.