US, China urged to cooperate on AI safety risks

What happened

A major artificial intelligence conference in Beijing's Zhongguancun district brought together computing pioneers and AI researchers, including speakers on recursive self-improvement and humanoid robots. Participants, including MIT computer scientist Stephen Casper, emphasized that the US and China should set aside competitive tensions to address shared AI cybersecurity and systemic risks.

Why it matters

The US currently restricts Chinese access to advanced AI chips and has ordered companies like Anthropic to block foreign nationals from its most powerful models over national security concerns. However, conference speakers argued that both nations stand to lose if AI develops too quickly without safety safeguards—citing vulnerabilities in AI-generated code, attacks enabled by agentic AI tool use, and automated social engineering as universal threats that transcend borders.

What to watch

Chinese companies have taken the lead in releasing highly capable open-weight models like Moonshot's Kimi, Alibaba's Qwen, and Z.ai's GLM, while the US has rebooted efforts with models like Nvidia's Nemotron. The field is approaching an inflection point where even less powerful open models could become dangerous if stripped of safety guardrails—and sources suggest some Chinese AI companies are already stepping back from open-source releases due to security concerns.