Anthropic's next-gen AI model "Claude Mythos," internally codenamed "Capybara," is reportedly in secret testing. Leaked documents indicate it represents a major leap in capabilities, surpassing the current top model Claude Opus. However, Anthropic has raised unprecedented cybersecurity risk warnings about the model, highlighting the tension between advancing AI power and ensuring safety. Its emergence is set to intensify industry competition and redefine the benchmarks for large language models.