GLM-5.3 reportedly approaches Claude Mythos Preview in exploit-writing tests

A post summarizes Anthropic’s Frontier Red Team report on Z.ai’s open-weight GLM-5.3. In tests targeting Chrome’s V8 engine, GLM-5.3 produced working exploit code in 50 of 410 attempts, compared with 56 for Claude Mythos Preview. The post also says Anthropic found that simple methods bypassed GLM-5.3’s refusals 64–100% of the time in its simulations.