Zhipu's GLM-5.3 matches Claude Mythos Preview at building exploits

zhipu's glm-5.3, an open-weight model, has been analyzed by anthropic's frontier red team and found to be nearly on par with claude mythos preview in building cyber exploits. glm-5.3 can build complete cyber exploits on its own, with anthropic's tests showing it can take full control of a target program in 4% of tasks. when paired with a human expert, glm-5.3 found previously unknown vulnerabilities in a widely used browser's javascript engine and chained them into a working attack. anthropic's analysis also showed that glm-5.3's safeguards can be easily bypassed using simple methods. the model's ability to build exploits at a low cost, with one test estimating a cost of $20.40, raises concerns about the potential misuse of open-weight models.