The model is GLM-5.3, built by Zhipu AI, or Z.ai.
Anthropic tested it against known vulnerabilities in Chrome’s V8 JavaScript engine.
Out of 410 attempts, GLM-5.3 produced 50 complete exploits.
Anthropic’s Claude Mythos Preview managed 56.
Mythos is restricted to vetted cybersecurity researchers.
GLM-5.3 is open-weight. Anyone can download it.
Anthropic also found its safeguards were relatively easy to remove. One technique reportedly bypassed them 100% of the time, with little loss in capability.
And stripped versions of the model appeared online within days.
In controlled tests, Anthropic says GLM-5.3 even discovered previously unknown browser vulnerabilities and chained them together.
Source: SCMP