Anthropic accidentally exposed draft blog post materials in an unsecured public data lake, revealing the existence of a new model codenamed "Mythos" that has been internally described as a "step change" in performance. The leak included nearly 3,000 unpublished assets linked to Anthropic's blog infrastructure. While the company has since acknowledged that it is testing the Mythos model, Anthropic has declined to provide a timeline for public release. The discovery underscores both the intensity of AI development competition and the operational security challenges facing frontier AI labs during periods of rapid innovation.