GLM-5.3 and the spread of advanced cyber capabilities

Zhipu AI’s GLM-5.3 can autonomously build end-to-end cyber exploits, achieving 12% success on ExploitBench (50/410) and 4% full control-flow hijacks on a 100-task OSS-Fuzz benchmark. Attackers bypass its safeguards 64–100% of the time in simulated tests, while safeguarded Claude models fail to do so. NIST’s CAISI assessment labels GLM-5.3 the most cyber-capable open-weight model, lagging U.S. frontier models by about four months.