We have a year to fix security everywhere

Cheap, unrestricted AI models like GLM 5.3-flash are now accessible to anyone and capable of assisting with dangerous hacking and illegal activities, creating an urgent security threat. The author argues that the tech industry has approximately one year to identify and fix vulnerabilities across systems before these models become widely deployed, though the main challenge lies in actually deploying those fixes rather than discovering them. Advanced AI systems can now find security issues faster than humans, presenting a rare opportunity to address vulnerabilities at scale if coordinated action is taken quickly.