Anthropic's 'Most Capable' AI Model Claude Mythos Leaks, Deemed Major Cybersecurity Threat
Anthropic is developing a new AI model called Claude Mythos, described as its most advanced and powerful system to date. Internal materials indicate Mythos significantly outperforms previous models, particularly in reasoning, coding, and cybersecurity. A leak of draft materials via an unsecured content management system exposed details, prompting Anthropic to restrict access and attribute the breach to human error. The model is undergoing early-access testing with select organizations, especially those focused on cybersecurity, due to concerns over its ability to exploit vulnerabilities more effectively than existing tools. The draft also refers to a future, even more advanced version called “Capybara.” News of the leak led to a selloff in cybersecurity stocks, with major companies’ shares dropping, reflecting market concerns over AI’s potential to disrupt enterprise software sectors. Previous Anthropic announcements, such as Claude Cowork, have similarly triggered significant market reactions as investors reconsider foundational AI’s ability to compete with established software providers.
