China May Have Just 'Reset' the AI Race

A Chinese startup's new AI model has Silicon Valley buzzing, reports CNBC. Cybersecurity researchers say the model from Beijing-based Zhipu AI now rivals Anthropic's specialized Mythos system at spotting software flaws—a key capability in defending, or attacking, critical networks. The Wall Street Journal suggests the development is "poised to reset the global tech race." The model, GLM-5.2, is "open-weight," meaning anyone can download, run, and modify it, making it attractive both to cost-conscious companies and to hackers who want to operate out of sight. It has already cracked the top 10 most-used models on aggregator OpenRouter, and in some tests has outperformed Anthropic's Claude Opus.

China's 360 Security just introduced its own Mythos-like bug-hunter, and its CEO framed such tools as too strategically important to leave "solely in American hands." The timing is tricky for the US, where the Trump administration has restricted access to several advanced Anthropic and OpenAI models over security fears, even as it allows AI chip exports to China. Critics warn that clamping down on US systems while Chinese open-weight models spread worldwide risks weakening American cyber defenses and accelerating China's rise in the AI race.