ChainThink reports that on March 27, 2026, according to 1M AI News, Anthropic inadvertently leaked the Claude Mythos model due to a misconfiguration in its content management system, which stored unpublished blog drafts and other assets related to the model in an unencrypted, publicly searchable database.
Claude Mythos is a new model currently being tested by Anthropic, positioned above the existing Opus series. Anthropic has confirmed that it represents a leap in capability and is the strongest model built to date. Leaked drafts show that the model achieves significantly higher scores than Claude Opus 4.6 in tests involving software programming, academic reasoning, and cybersecurity, with cybersecurity performance far surpassing all existing AI models. The drafts also mention the codename "Capybara," which points to the same underlying model as Mythos.
The leak also involved nearly 3,000 unpublished assets, including images, PDFs, and details of an exclusive closed-door summit for European corporate CEOs to be held in the UK, which CEO Dario Amodei will attend. The incident was independently discovered by Alexandre Pauwels, a cybersecurity researcher at the University of Cambridge, and Roy Paz, Senior AI Security Researcher at LayerX Security. Anthropic stated that due to the model’s considerable capabilities, it will adopt a cautious release strategy and will not be made publicly available in the short term; instead, it will initially be offered to a small group of early customers, with a focus on cybersecurity defense use cases, as the model incurs extremely high operational costs.