How Comcast and Booz Allen use Claude Mythos to find exploit chains and secure their codebases

Anthropic’s Project Glasswing, a limited‑access program, now offers its Cyber Verification Program to security professionals, providing advanced AI models such as Claude Mythos to identify vulnerabilities that conventional scanners overlook. Comcast, a global media and technology company, employed Claude Mythos Preview to assess 258 business‑critical systems comprising approximately 170 million lines of code. The model uncovered a critical authentication vulnerability that could have allowed attackers to bypass controls; the issue was remediated before any exploitation was observed. The vulnerability arose from interactions among multiple systems rather than a single defective component, illustrating the model’s capacity to maintain context across code, configuration, and application behavior. Booz Allen, a defense technology firm, used the same model to review a broader software portfolio within a fixed engagement window. The model identified a flaw in device‑startup code that left a security key unprotected, enabling an attacker to prevent the lock‑or‑erase step and keep a compromised device accessible. Both organizations reported that Claude Mythos accelerated discovery, enabled rapid remediation, and highlighted the need for continuous security reviews to keep pace with AI‑enabled adversaries. Anthropic plans to expand access through the Cyber Verification Program, aiming to give more defenders the ability to examine systems from an attacker’s perspective and close gaps before exploitation occurs.