Anthropic's Secret 'Mythos' AI — A Data Leak Reveals the Most Powerful Model Ever Built

Fresh details are pushing this story further into focus. Anthropic's most powerful AI system to date has been inadvertently revealed — not through a planned announcement, but through a data leak that exposed draft blog posts, product plans, and confidential documents stored in an unsecured, publicly searchable data cache. The model is called Claude Mythos., Anthropic describes it as 'by far the most powerful AI model we've ever developed' and a 'step change' above its current flagship, Claude Opus 4.6. The company has confirmed the model's existence. The leaked documents describe a new model tier called 'Capybara' — positioned above Opus as the highest performance bracket. Capybara and Mythos appear to refer to the same underlying system. Most alarmingly, Anthropic's own internal documents flag the model as 'currently far ahead of any other AI model in cyber capabilities' — representing a risk that could help hackers mount large-scale, automated cyberattacks at a scale that outpaces defenders. The company says it is rolling out access specifically to cybersecurity organisations first, to give defenders a head start. The leak was discovered by cybersecurity researcher Roy Paz of LayerX Security and Alexandre Pauwels of the University of Cambridge. Anthropic attributed the exposure to 'human error' in configuring its content management system and has since secured the cache. Also revealed: an invite-only CEO summit at an 18th-century English manor, where Anthropic CEO Dario Amodei plans to showcase unreleased Claude capabilities to Europe's most influential business leaders. The picture now is one of rising pressure, wider consequences and very little room for error.