A data leak has revealed that Anthropic is developing "Claude Mythos," its most powerful AI model to date, currently in early testing. Internal documents, accidentally exposed online, detail this new model, which is described as a new tier named "Capybara," surpassing their current Opus models in intelligence and size, and consequently, cost. The leak also highlighted significant cybersecurity risks associated with Claude Mythos, as it could potentially exploit software vulnerabilities faster than existing tools, leading to more sophisticated cyberattacks. Anthropic attributes the leak to human error in their content management system and plans a cautious, early-access release to trusted organizations to help them prepare for AI-driven exploits, having already observed misuse attempts by hacking groups.