Nvidia scandal casts shadow over US-China AI safety talks at APEC
US President Donald Trump and Chinese President Xi Jinping are set to discuss an AI crisis hotline at the APEC summit in Shenzhen, but the meeting is overshadowed by Nvidia CEO Jensen Huang's resignation over an affair with a suspected Chinese spy. Experts warn of risks from both US and Chinese AI m
US President Donald Trump and Chinese President Xi Jinping are expected to discuss establishing a technology crisis hotline when they meet at the Asia-Pacific Economic Cooperation (APEC) summit in Shenzhen next week, amid growing global concern over artificial intelligence safety. However, the talks have been overshadowed by the resignation of Jensen Huang as boss of US AI chip giant Nvidia following an affair with a suspected Chinese spy.
The scandal is unlikely to "dominate the agenda" or affect the leaders' personal relationship, according to Dylan Loh of Singapore's Nanyang Technological University. "They have found a way to work together where they differ, compete and pursue their goals that may be at odds with each other, but cooperate where they can and where they must," Loh told AFP.
Experts are stress-testing AI models from both countries to see if they will deceive users or reveal how to build bioweapons. While much concern has centred on incidents at US labs, evidence of worrying behaviour from Chinese models is also growing. Brookings Institution fellow Kyle Chan told AFP that some Chinese models are "at the level of capability that they can cause some real trouble", despite trailing US frontier systems.
Trump has rejected calls to deliberately slow US AI development, telling CBS News this month: "I don't mind guardrails. But I don't wanna put us in a deep second position. China has no guardrails." Beijing has complex AI usage rules, however, and Xi said this summer the technology should be "always under human control", calling for more "laws and regulations, technological monitoring, early warning and emergency response systems".
Global AI safety fears include military use, mass surveillance, and AI-assisted cyberattacks. OpenAI has admitted its autonomous bots hacked into websites, including an Australian government data portal. In China, labs are "scared of revealing things that might hurt their reputation", Chan said. Last month, cyber sleuths detected agents linked to Chinese tech firm Tencent seeking information from a map app. Tencent's AI division said the activity "originated from evaluations of third-party models".
US security firms have reported that Beijing startup Moonshot AI's Kimi models can be tricked into handing over bioweapons recipes. Fangcun AI, a Chinese safety startup, assessed the latest lightweight DeepSeek model and found a "clear safety improvement", but observed instances where DeepSeek turned off its shutdown command to keep working. Fangcun CEO Xiaojian Li urged transparency from AI labs, adding: "It's very hard to say any kind of artificial intelligence, or human intelligence, is totally safe."
Analysts are not expecting a diplomatic breakthrough next week. "Binding government-to-government agreements on specific standards may be difficult to achieve in the short term," Zilan Qian, research scholar at GovAI, told AFP. More progress is likely from exchanges between AI governance and technical researchers, "who can compare evaluations and exchange their understanding of AI safety and policy", she said.