Major Chinese AI firms curb new models’ cyberattack capabilities
Reported by Bloomberg, Startup Fortune, 联合早报 and 1 more outlets
Moonshot AI, Tencent and Alibaba have released models with reduced offensive cyber capabilities, while Zhipu plans to restrict its most sensitive functions to verified users. The moves follow a Chinese government proposal to restrict exports of such technology.
Key points
- Moonshot, Tencent and Alibaba released models with reduced offensive cyber capabilities
- Moonshot says its filtered model solved 81% of tasks on its exploit benchmark, versus 90.5% for an unfiltered build
- Zhipu plans to put its most sensitive capabilities behind a verification process; it has not announced when it will open
Sources
- BloombergRead ↗China’s Top AI Firms Limit Cyber Capabilities in Latest Models
- Startup FortuneRead ↗Moonshot, Tencent and Zhipu trim hacking skills from new Chinese AI models
- 联合早报Read ↗Major Chinese AI firms curb new models’ cyberattack capabilities
- AP NewsRead ↗Chinese tech firms fear AI hits from US curbs after Nvidia fallout
Details
Startup Fortune reports that Moonshot AI’s Kimi K3.1 Pro, Tencent’s Hy4 and Alibaba’s Qwen4-Max have reduced offensive cyber capabilities. Tencent’s Hy4 is open source; Qwen4-Max was released without open weights.
Moonshot says it removed cyber content and vulnerability-research tasks from Kimi K3.1 Pro’s training data and reinforcement-learning environments while preserving general coding. On Moonshot’s own, non-public xdev-bench-hard benchmark, the filtered model scored 81.0%, compared with 90.5% for an unfiltered build; the figures have not been independently checked, according to Startup Fortune.
Zhipu plans to reserve its most sensitive functions for verified users through a program called Cybersecurity Trusted Access, rather than removing those capabilities.
Written from coverage as of
Images
- Startup Fortune
Stocks in this story
Change since open: 20 Nov to 21 Nov 2026
Last refresh