China weighs open-weight AI security risks against tech innovation strategy

Experts in policy are warning that China might tighten its rules on open-weight AI models as they start to match the performance and risks of top-tier models. Traditionally, these open-weight models lagged behind proprietary versions by several months. However, recent breakthroughs from Chinese firms, like Zhipu AI's GLM-5.2, have significantly narrowed this gap. Mark Witzke, a non-resident scholar at UC San Diego who focuses on US-China tech policy, noted that if these open-weight models begin to present cybersecurity and biosecurity threats akin to frontier models, China may deem them too dangerous for public use, mirroring concerns in the US. Such a regulatory shift could impact China's approach to technological innovation.