China narrows AI gap with US as both powers weigh risks
Chinese AI models are closing in on US frontier systems even as both governments flag risks, from cyber threats to model distillation.
China is rapidly closing the distance to the United States in artificial intelligence, even as both countries wrestle publicly with the dangers the technology poses. American models have led for years in capability, research output and citations, but recent assessments suggest that advantage is thinning.
A Stanford Institute for Human-Centered Artificial Intelligence report and independent experts describe China as an emerging AI counterweight. Samm Sacks of Johns Hopkins University's School of Advanced International Studies called the gap between US and Chinese models "narrow and fragile."
The stakes are heightened by an expected visit from Chinese President Xi Jinping to Washington this month for talks with US President Donald Trump. Analysts say the trajectory of AI will hinge heavily on decisions made in both capitals.
Several Chinese systems have drawn attention. Beijing startup Moonshot AI's Kimi K3 briefly ranked third globally in model intelligence in July, behind Anthropic's Claude Fable 5 and OpenAI's GPT-5.6 Sol, before slipping to ninth as rivals released new frontier models. DeepSeek's R1, launched in January 2025, was the first to unsettle global markets by showing Chinese models could compete on cost-effectiveness; the company previewed a more capable V4 in April. Z.ai's GLM-5.2 and GLM-5.3 followed, earning comparisons to top US systems, while Alibaba previewed its Qwen3.8-Max in July.
Concerns are not one-sided. China's state security minister, Chen Yixin, warned in a Sunday article that AI could threaten political, institutional and ideological security if misused by hostile actors, citing risks such as deepfake-driven political rumors and AI-assisted identification of cybersecurity vulnerabilities. He pointed to the capabilities of Anthropic's Claude and OpenAI's GPT as examples of how hacking tasks could become easier. Chinese leaders have simultaneously called for faster AI development, technological self-reliance and a global governance framework.
Washington has accused Chinese developers of "industrial-scale distillation," alleging they extract capabilities from advanced US models. On Sept. 8, the FBI, National Security Agency and Cybersecurity and Infrastructure Security Agency issued their most comprehensive report to date, describing "aggressive, malicious, and targeted distillation activities." Treasury Secretary Scott Bessent said the same day that China distills US models and "can never get ahead."
Beijing rejected the claims, noting distillation is a common industry technique, and urged the US to avoid "unfounded accusations" while accusing Washington of seeking an AI monopoly. Lizzi C. Lee of the Asia Society Policy Institute's Center for China Analysis said the allegations remain contested but carry strategic weight, as the US shifts from controlling AI inputs like chips to scrutinizing outputs as a form of technology transfer.