Anthropic Launches Mythos User-Sharing Network for Threat Intelligence

Golden Finance reports that on May 19, out of concern that limited access to information could harm the interests of small and medium-sized enterprises, Anthropic recently revised its original position and officially allowed users of its high-performance AI model Mythos to share cybersecurity threat intelligence with other organizations that face similar security-vulnerability risks. This new rule highlights the real-world dilemma that many leading AI companies encounter when tightly controlling access to top-tier models. Previously, Anthropic—through a project called the Glass Wing Program—only granted Mythos access to about 50 large enterprises and institutions responsible for core digital infrastructure. The model’s efficiency at discovering software vulnerabilities far surpasses manual inspection. The project’s original intent was to enable these enterprises and government agencies to address cybersecurity risks in advance, and then gradually expand the rollout of access and introduce more powerful AI models. An Anthropic spokesperson said that the first batch of Mythos users had previously signed confidentiality agreements strictly prohibiting the external disclosure of the cybersecurity risk information they had. Last week, Anthropic officially notified its partner companies that, while adhering to the principles of compliance and prudence, they may exchange network threat intelligence and the security-issue findings produced by the Mythos model with other institutions.