China faces a regulatory balancing act between the security risks of open-weight AI models and an innovation strategy crucial to its technological race with the US, researchers said. Open-weight models, which allow free code downloads for local hardware, have historically lagged behind proprietary frontier models, but recent Chinese lab releases have narrowed that gap. Zhipu AI’s GLM-5.2 became the first Chinese large language model (LLM) to rank among the top three globally on leading benchmarks, earning praise as the country’s first open-weight model reliable for daily coding workflows. As these models advance, analysts warned of potential regulatory responses. Mark Witzke, a non-resident scholar at the University of California San Diego, stated that as open-weight models approach the cyber and biosecurity risks of Mythos and other leading-edge models, China may make the same calculation as the US and find them too dangerous to release in open form. The concerns stem from Anthropic’s Claude Mythos, announced in April, which rattled global industries by autonomously identifying and exploiting cybersecurity vulnerabilities. Amid warnings that its reasoning could lower technical barriers to developing bioweapons, the US lab restricted Mythos to a select group of American organisations.