Anthropic accidentally leaks unreleased Claude Mythos model

Anthropic’s upcoming model Claude Mythos was inadvertently exposed via a publicly accessible data cache due to human error, along with details of a new higher-tier Capybara model positioned above Claude Opus. The leak revealed cybersecurity risk assessments and model specifications. This matters because premature disclosure of frontier model capabilities can influence competitive dynamics, investor perceptions, and safety evaluations across the AI industry.