I Met With China's Top AI Experts. They're Freaking Out, Too

Researchers at a major AI conference in Beijing argued that the United States and China must set aside their technological rivalry to jointly address the escalating cybersecurity and systemic risks posed by increasingly capable frontier artificial intelligence models. The conference, organized by the Beijing Academy of Artificial Intelligence (BAAI) [7] in the Zhongguancun district, featured sessions on recursive self-improvement and humanoid robots, and included computing pioneers Whitfield Diffie and Andrew Barto [1]. Stephen Casper, a computer scientist at MIT, stated via video link that “AI is a global technology with global benefits, global harms, and a consistent tendency for new capabilities to eventually proliferate” [1]. He added that “one thing that almost everyone in AI can agree on right now is that AI doesn't need a Chernobyl moment” [1]. The call for cooperation comes amid heightened tensions. Washington has imposed tight restrictions on chips and chipmaking equipment to slow China’s AI progress, and recently ordered Anthropic to block foreign nationals from accessing its most powerful models over national security concerns [1]. Nvidia, which as of 2025 controlled more than 80% of the market for GPUs used in training and deploying AI models [2], is central to this hardware competition. Casper pointed to research indicating the benefits of international collaboration on AI dangers outweigh the national security risks of working together, comparing the dynamic to Cold War-era nuclear cooperation between the US and the Soviet Union [1]. Lin Yun, a professor at Shanghai Jia Tong University, said hackers would gain an advantage in the near term, but that new countermeasures should eventually tip the balance back toward defense [1]. “If different countries understand the risks in similar ways, it becomes easier to develop shared safety principles and technical standards,” Yun said. “The key is to find areas where sharing can reduce systemic risk without exposing sensitive operational details” [1]. A central tension is the role of open-weight models. Chinese companies have led in releasing highly capable open-weight AI, such as Z.ai’s GLM 5.2, which includes frontier agentic and coding capabilities [1]. The BAAI itself frequently releases new models and open-source code as part of its mission to promote collaboration between academia and industry [7]. However, Yun noted the industry must devise ways to guarantee open models are up-to-date, free of backdoors, and meet safety standards [1]. A source at a leading Chinese AI company, speaking anonymously, indicated that security concerns are already prompting some advanced models in China to no longer be released as open source [1].