China's Z.ai says new model nears Anthropic's Mythos 5 in cyber-defence tests

Chinese AI startup Z.ai announced on Friday that its open-source GLM-5.3 model was nearing Anthropic's restricted Mythos 5 in cybersecurity tests. The GLM-5.3 model scored 84.5% on CyberGym, a test measuring a model's ability to review code, identify security flaws, and confirm their authenticity. This was slightly higher than Mythos 5's 83.8% score.

However, GLM-5.3 lagged behind Mythos 5 in converting discovered flaws into working attacks, with a 54.4% score compared to Mythos 5's 78.0% on the ExploitBench test. In a timed test, GLM-5.3 completed 105 attack-development tasks in two hours and 130 in six hours, while Mythos 5 completed 181 and 247 tasks, respectively. Anthropic has limited access to its Mythos model to vetted organizations due to concerns about potential misuse.

Z.ai plans to release GLM-5.3 publicly in two weeks after completing security assessments and enhancing its safeguards, with sensitive cybersecurity functions available only to verified users through a trusted access program. This move reflects China's growing sophistication in open-weight risk management practices, according to AI governance researcher Gabriel Wagner.