Summary
Z.ai said GLM-5.3 reached unusually strong performance on vulnerability discovery and exploit-development benchmarks. The company will initially provide controlled access to selected security partners and delay public weights for two weeks.
Why it matters
Open-weight models are approaching frontier cyber capability while remaining harder to control after release. The decision also shows that release governance is becoming a competitive issue between Chinese and US labs.
Related coverage
- Z.ai opens vulnerability scanning to open-source maintainers
- Cyber risk is changing frontier-model release schedules
- Anthropic raises its assessed risk for high-stakes misalignment