Currently, the company is exclusively releasing Project Glasswing to “defensive security” partners. Anthropic is introducing a new AI model as part of a cybersecurity collaboration with Nvidia, Google, Amazon Web Services, Apple, Microsoft, and others. Known as Project Glasswing, it aims to enable large corporations and potentially the government to identify system vulnerabilities with minimal human input.
Anthropic is providing its launch partners access to Claude Mythos Preview, a general-purpose model not planned for public release due to security issues. Newton Cheng, cyber lead for Anthropic’s red team, stated that the model offers cyber defenders a “head start” against threats. Partners will utilize the model to identify critical vulnerabilities and assist in their resolution. Access is restricted to prevent adversaries from exploiting it.
Although Claude Mythos Preview wasn’t specifically developed for cybersecurity, Anthropic attributes its advances to its “strong agentic coding and reasoning skills.” In an interview, Cheng refrained from detailing the model’s successes beyond publicly available examples, but Anthropic’s blog noted that Mythos Preview has identified “thousands of high-severity vulnerabilities” recently. The blog underscores that the model autonomously identified vulnerabilities and related exploits without human guidance.
Claude Mythos Preview’s existence was first revealed last month due to a data leak, attributed to human error by Anthropic. Dianne Penn, a product management head, emphasized their commitment to process improvements.
Mythos Preview will be privately accessible to partners such as JPMorgan Chase, Broadcom, Cisco, CrowdStrike, the Linux Foundation, and Palo Alto Networks, among others. Anthropic plans to subsidize its usage costs, committing up to $100 million in credits and $4 million in donations to the Linux Foundation and Apache Software Foundation. Eventually, the program could evolve into a revenue-generating service.
Despite a recent public dispute with the Trump administration, Anthropic disclosed ongoing talks with US officials about Mythos Preview’s capabilities. Penn confirmed briefings with senior US government officials, reiterating Anthropic’s commitment to governmental collaboration, though Cheng did not specify the officials involved.