Anthropic states Zhipu's open-weight GLM-5.3 nearly matches Claude Mythos Preview in exploit development

Anthropic states Zhipu AI's open-weight model GLM-5.3 nearly matches Claude Mythos Preview in building cyber exploits without effective safeguards. On ExploitBench testing Chrome V8 bugs, GLM-5.3 built working exploits in 50 of 410 attempts versus 56 for Mythos Preview. Furthermore, a smaller Flash variant created a reliable Chrome attack for $20.40 at API prices, and the US agency CAISI supports these findings.