Z.ai's GLM-5.2 Matches US Models on Security Bugs
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- WSJ reports Z.ai's GLM-5.2 matches the latest US models at finding security bugs, arguing that US clampdowns on top AI models may be "handing Beijing a cyberwarfare advantage"
- Jefferies (via Business Today and China Money Network) describes China's cheaper AI models as "quietly taking over enterprise AI," with Z.ai's pricing—not just intelligence—as the central competitive lever
- Aaron Levie (Box CEO) predicts mythos-level cybersecurity models will soon be open and available to anyone, driving "economic value and control away from the US's tech stack"
- Kyle Russell shared internal benchmarking showing GLM 5.2 was "a bit better at less than half the cost" of Opus 4.8 for updating his firm's internal mortgage servicing knowledge base
- Tim Hua publicly challenged the WSJ framing, offering $100-to-$1 odds that GLM-5.2 will score below Mythos and GPT-5.5 at UK AISI's cyber range
- Koen Bok called GLM 5.2 running at 200tps combined with US restrictions on frontier model rollouts "10x bigger than DeepSeek 2025"
- Guillermo Rauch warned that Mythos/Sol-class cybersecurity capabilities are equally useful offensively, posing "a serious threat to US companies that remain unaware of latent vulnerabilities"
Why it matters: Kyle Russell's firm found GLM 5.2 'a bit better at less than half the cost' of Opus 4.8 for real knowledge-base work, and Koen Bok notes it runs at 200tps—making enterprise adoption of Chinese open-weight models a live cost question, while Guillermo Rauch warns the same cyber capabilities could be weaponized against US firms.
