I met the best AI experts in China. They’re terrified, too

Just over a A week ago, I attended a major artificial intelligence The conference was held in Zhongguancun, Beijing’s bustling high-tech district.

It was full of fascinating sessions that touched on everything from iterative self-improvement — the idea that models can modify their code and progress indefinitely — to humanoid robots. It featured a few computing legends, including Whitfield Diffie, co-inventor of public-key cryptography, and Andrew Barto, who won a Turing Award with Rich Sutton for his pioneering work in reinforcement learning.

But I left with one conclusion above all: The United States and China should put aside their fierce competition in artificial intelligence.

The cybersecurity and systemic risks facing Frontier AI are too serious to ignore, and increasingly capable client models could soon wreak havoc unless the world’s AI superpowers can work together. “AI is a global technology with global benefits, global harms, and a consistent tendency for new capabilities to eventually spread.” Stephen CasperThe MIT computer scientist who spoke on the videoconference told me afterward.

Until now, the United States has viewed China’s progress in artificial intelligence as an economic and national security threat. Washington has imposed tight restrictions on chips and chip manufacturing equipment to thwart the country’s development of powerful artificial intelligence. The latest is the American government commander Humanitarian to prevent foreign citizens from accessing his most powerful models, the Mythos and Fable 5, is over National security concerns. In response, Anthropic has revoked access for everyone. One company that was particularly concerning, as WIRED previously revealed, was The South Korean telecom giant with alleged ties to China.

But the conference, organized by the Beijing Academy of Artificial Intelligence, reinforced the idea that the United States and China stand to lose if AI is developed too quickly and recklessly. As artificial intelligence becomes more powerful, more effective, and more intertwined with everyday life, the risks that it can be used to launch cyberattacks or fail in catastrophic ways will increase. Since the world’s two dominant AI powers are responsible for the most advanced models, it seems that collaboration between them will be crucial.

Casper pointed out research It shows that the benefits of international cooperation on AI risks outweigh any national security risks that come from working together. He likened the current situation to how the United States and the Soviet Union were forced to work together on nuclear risks, even as each sought to outdo the other in stockpiling nuclear weapons.

“The one thing that almost everyone in AI can agree on right now is that AI doesn’t need a Chernobyl moment,” Kasper said.

One day-long session highlighted the universality of the cyber challenges raised by more advanced artificial intelligence. This includes new types of vulnerabilities in Code generated by artificial intelligenceand new ways to attack enabled systems through the use of proxies and automated methods of execution Social engineering Attacks.

After another session, I spoke with Lin Yun, a professor at Shanghai Jia Tong University He does an excellent job On artificial intelligence and computer security. Yoon told me that he expects hackers to gain an advantage in the near term, but new countermeasures, including new uses of artificial intelligence, should tip the balance back toward defense over time.

Even if international cooperation is complicated by competition, it should remain a priority, Yun said. “If different countries understand risks in similar ways, it becomes easier to develop common safety principles and technical standards,” Li said. “The key is to find areas where sharing can reduce systemic risk without revealing sensitive operational details.”

Perhaps the most pressing question for both countries is how to balance openness and risk. Open-weight models have become increasingly important for research and innovation, with Chinese models proving popular in the United States. But as these models advance, it will become more difficult to ensure that they do not help hackers identify security vulnerabilities and cannot be used as cyber weapons.