The GLM-5.3 model, introduced by Zhipu, has achieved notable advancements in its cybersecurity prowess. It is now not only capable of replicating vulnerabilities but also adept at further exploiting them. Nevertheless, these high-risk functionalities have been meticulously segregated and controlled, ensuring they remain inaccessible to the general public for now. Presently, GLM-5.3 has been integrated into the GLM Coding Plan, ZCode, and AutoClaw. Tasks involving vulnerability exploitation and attack verification are currently entrusted solely to controlled partners for rigorous testing. Subsequently, these capabilities will be extended to verified users via the 'Cybersecurity Trusted Access' program. The complete model weights will be released only after a thorough security assessment has been conducted.