Anthropic has inadvertently leaked its upcoming Claude Mythos model, revealing it outperforms the existing Opus series, particularly in cybersecurity. The leak occurred due to a misconfiguration in Anthropic's content management system, exposing unpublished drafts and assets in an unencrypted database. The Claude Mythos model, codenamed "Capybara," demonstrates significantly higher scores in software programming, academic reasoning, and cybersecurity compared to Claude Opus 4.6. The leak, discovered by cybersecurity experts Alexandre Pauwels and Roy Paz, also included details of a closed-door summit for European corporate CEOs. Anthropic confirmed the model's advanced capabilities and plans a cautious release strategy, initially offering it to select early customers focused on cybersecurity defense, citing high operational costs.