I Spoke with Leading AI Experts in China, and They’re Concerned as Well.

Just over a week ago, I participated in a significant AI conference held in Zhongguancun, a vibrant high-tech hub in Beijing.

The event featured an array of intriguing sessions covering topics from recursive self-improvement—the concept that models can modify their own code and evolve without limit—to humanoid robotics. It also included notable figures in computing, such as Whitfield Diffie, co-inventor of public-key cryptography, and Andrew Barto, who, alongside Rich Sutton, received the Turing Award for groundbreaking work in reinforcement learning.

However, I walked away with one key insight: The intense rivalry between the US and China over AI should be reconsidered.

The cybersecurity and systemic risks posed by frontier AI are too significant to overlook, and the increasing capabilities of agentic models could lead to chaos unless the leading AI nations collaborate. “AI is a global technology yielding global benefits and harms, with a consistent trend of new capabilities eventually spreading,” Stephen Casper, an MIT computer scientist who participated via video, remarked to me afterwards.

Historically, the US has regarded China’s advancements in AI as threats to its economic and national security. Washington has implemented stringent controls on chips and manufacturing equipment to hinder China’s development of powerful AI technologies. Recently, the US government directed Anthropic to restrict access to its most advanced models, Mythos and Fable 5, due to national security issues, leading Anthropic to revoke access for all users. A particular concern was a South Korean telecom giant allegedly linked to China, as previously reported by WIRED.

Yet, the conference, hosted by the Beijing Academy of Artificial Intelligence, underscored that both the US and China risk significant losses if AI is advanced in a reckless manner. As AI evolves to become more powerful, agentic, and integrated into daily life, the likelihood of it being used for cyberattacks or facing catastrophic failures will increase. Given that these two nations lead in AI model development, their cooperation seems essential.

Casper highlighted research showing that the advantages of international collaboration in addressing AI threats surpass any national security concerns arising from such partnerships. He compared the current scenario to the necessity of US-Soviet cooperation on nuclear threats, even amid an arms race.

“Nearly everyone in AI agrees at present that we should avoid a Chernobyl moment,” Casper stated.

A full-day session emphasized the common cyber challenges presented by more sophisticated AI. This includes emergent vulnerabilities in AI-generated code, innovative methods of attacking systems using agentic tools, and automated strategies for conducting social engineering attacks.

Following another session, I chatted with Lin Yun, a professor at Shanghai Jia Tong University, renowned for his work on AI and cybersecurity. Yun conveyed his belief that hackers will gain an upper hand in the short term, but that innovative countermeasures, including new applications for AI, should eventually restore the advantage to defensive strategies.

Yun emphasized that despite the complexities introduced by competition, international collaboration must remain a priority. “When different countries share a common understanding of the risks, it becomes easier to establish unified safety principles and technical standards,” he noted. “The challenge lies in identifying areas where cooperation can mitigate systemic risk without exposing sensitive operational information.”

A pressing dilemma for both nations is how to balance transparency with safety. Open-weight models have become essential for research and innovation, with some Chinese models gaining traction in the US. However, as these models progress, ensuring they do not aid hackers in discovering security flaws or becoming tools for cyberattacks will become increasingly difficult.