BEIJING — The conference hall in Zhongguancun was full of people who build the future. Whitfield Diffie, the man who helped invent public-key cryptography, was there. Andrew Barto, a Turing Award winner for reinforcement learning, was in the room.
But the talk kept circling back to what could break. For three days, Chinese AI researchers and visiting experts from the West debated recursive self-improvement and humanoid robots.
The sessions were packed. The real story, though, was not about any new model or breakthrough. It was about a shared fear that the US and China are racing toward a cliff together, each convinced the other is the problem.
Stephen Casper, a computer scientist at MIT, spoke via video. His message was blunt: “AI is a global technology with global benefits, global harms, and a consistent tendency for new capabilities to eventually proliferate.” He did not mean that in a hopeful way.
The cybersecurity risks are what have the researchers spooked. Agentic models — AI that can act in the digital world, not just answer questions — are getting better fast. They can write code, move money, alter permissions.
If they are not locked down, they can cause real damage. One model trained in Shanghai can be downloaded in São Paulo and deployed against a hospital in Ohio. Borders mean nothing.
Until now, the US has treated the AI race as a national security problem. Export controls, chip bans, investment restrictions.
The message from Beijing was different. The researchers there argue that the threat does not respect those lines. A vulnerability in a foundation model is a global vulnerability.
A system that can tweak its own code — recursive self-improvement — is not something one country can contain. The economic stakes are enormous.
Everyone knows that. But the conference made a different calculation. The systemic risks are now too great for either side to ignore.
That was not a fringe opinion. It was the dominant takeaway from a room full of people who understand the technology best. What comes next is unclear.
The US government has shown little appetite for cooperation. National security hawks see any collaboration as a leak of advantage.
The Chinese researchers see it as survival. Neither side can secure a global technology alone. A model trained in one country can be deployed in another.
A vulnerability found in one system can be exploited everywhere. That is not a political statement.
It is an engineering fact. The conference did not produce a joint statement or a plan. It produced a warning.
The researchers who spoke were not activists. They were computer scientists, cryptographers, engineers. They see what is coming.
Agentic models are not a future problem. They are here.
And if the two superpowers cannot find a way to work together on security, the chaos will not stop at the border.
{“@context”: “https://schema.org”, “@type”: “NewsArticle”, “headline”: “Chinese AI Experts and Western Scientists Discuss AI Risks in Beijing Forum”, “description”: “BEIJING — The conference hall in Zhongguancun was full of people who build the future. Whitfield Diffie, the man who helped invent public-key cryptography, was…”, “datePublished”: “2026-06-24T16:40:00+00:00T09:30:00+00:00”, “author”: {“@type”: “Person”, “name”: “Arjun Mehta”}, “publisher”: {“@type”: “Organization”, “name”: “InfoPulse Today”, “url”: “https://newsfeed.infopulsetoday.com/”}}
