Anthropic Investigating Unauthorized Access to Cybersecurity Tool Mythos

Anthropic is currently investigating a security lapse involving Mythos, its high-stakes AI-powered cybersecurity tool. The breach, which allowed a group of unauthorized users to access the model, stems from a vulnerability within a third-party vendor environment. According to reports, these users successfully located the tool’s online access point shortly after its announcement, effectively bypassing restrictions meant to gatekeep the software. Mythos is a core component of 'Project Glasswing,' an initiative designed to provide advanced, controlled defensive capabilities to select corporate partners like Apple. Because the model is engineered to identify and potentially execute complex cyberattacks, Anthropic has long positioned it as a sensitive, high-risk asset. While the company maintains that its primary internal systems remain secure, the incident highlights a critical, often overlooked reality in the AI supply chain: even the most robust models are only as secure as the weakest link in their deployment pipeline. Whether the unauthorized users were motivated by curiosity or malice is secondary to the fact that a system designed for defense was exposed through a simple…