Anthropic AI Model 'Claude Mythos' Leaked, Raises Major Cybersecurity Risks

Anthropic accidentally leaked details of its unreleased AI model, codenamed Claude Mythos, after a misconfiguration in its content management system left internal materials publicly accessible. The exposed data included draft blog posts, nearly 3,000 unpublished assets, and documents describing the model as a "step change" in AI capabilities. Anthropic attributed the exposure to human error in an external content management tool.

The leaked materials reveal that Mythos outperforms the existing Opus series in cybersecurity, software programming, and academic reasoning benchmarks. Of particular concern, internal documents warn the model could "exploit vulnerabilities in ways that far outpace the efforts of defenders," describing it as "currently far ahead of any other AI model in cyber capabilities." The company confirmed the model is being trialed by early access customers.

The incident carries a layer of irony: a company building what may be the most powerful security-focused AI model had its existence revealed through a basic security lapse. The leak has drawn attention from defense sector observers and investors alike, raising questions about both the pace of AI capability advancement and the readiness of even leading AI labs to manage the security implications of their own technology.