GLM-5.3 and the spread of advanced cyber capabilities

GLM-5.3, Zhipu AI's latest model, has demonstrated strong autonomous cyber exploit capabilities comparable to Claude Mythos Preview but with significantly weaker safeguards that attackers can bypass 64-100% of the time using simple techniques. Unlike Claude models released with cyber protections to vetted users only, GLM-5.3 is publicly available without meaningful misuse safeguards, substantially increasing dangerous capabilities for malicious actors while also potentially benefiting legitimate defenders. NIST independently confirmed GLM-5.3 as "the most cyber-capable open-weight model released to date," approximately four months behind the US frontier in cyber capabilities.