Anthropic Limits Claude Mythos Release Over Cyber Risks

Anthropic has restricted the deployment of its Claude Mythos AI model following internal safety evaluations. The tests showed the system can autonomously find and exploit zero-day vulnerabilities in critical infrastructure ^[1]. Some of these security flaws remained undetected for nearly 30 years ^[1][2]. The company cited "catastrophic" cybersecurity risks as the reason for the limited release ^[2]. While some analysts view the model as a turning point for global security, others describe the announcement as part of a controlled preview known as "Project Glasswing" ^[1][3]. Experts warn that the ability of AI to target long-standing flaws creates new challenges for digital defense ^[2]. However, some industry observers suggest the restricted rollout may also serve as a strategic communication effort by the developer ^[3].