Jensen Huang Resigns as Head of U.S. AI Chipmaker Nvidia!

Jensen Huang

CHINA: AI bots going rogue and people using the technology to cause harm will likely be discussed when Donald Trump and Xi Jinping meet next week, with the Nvidia scandal affecting talks on AI safety.

The US and Chinese presidents are expected to discuss a tech crisis hotline at the Asia-Pacific Economic Cooperation (APEC) meeting in Shenzhen, China. Meanwhile, experts around the world are testing AI models to see if they might trick or disobey users, or reveal how to make a biological weapon.

Jensen Huang’s resignation as head of US AI chip company Nvidia, following an affair with someone suspected of spying for China, could distract from the talks.

But the scandal is unlikely to “take over the agenda” or affect Trump and Xi’s relationship, said Dylan Loh of Singapore’s Nanyang Technological University.

The leaders have found a way to work together “even where they disagree, compete and pursue goals that may conflict, while cooperating when they can and need to,” Loh told AFP.

Most worries about AI risks have come from incidents at US labs, but there is also growing evidence of concerning behaviour by Chinese models.

Some experts say Chinese AI is not yet advanced enough to cause serious concern about going rogue.

But although China is behind the most advanced US systems, some Chinese models are still “capable of causing real trouble,” Brookings Institution expert Kyle Chan told AFP.

Trump has repeatedly rejected calls to slow the development of America’s most powerful and potentially dangerous AI systems, arguing that doing so could help China get ahead in this important field.

“I don’t mind safety rules. But I don’t want us to fall far behind,” Trump told CBS News this month.

“China has no safety rules,” he said except that “whatever President Xi says goes.”

However, Beijing has detailed rules for how AI is used, and some Chinese companies are deliberately reducing their models’ ability to hack systems.

Xi said this summer that the technology should “always remain under human control.” He called for more “laws and rules, technology checks, early warnings and emergency response systems.”

Trump and Xi agreed in September to set up a way to quickly communicate about AI incidents. They are expected to work out the details at APEC on November 18–19.

Worldwide concerns about AI include its use by the military, mass surveillance and AI-assisted cyberattacks carried out by people.

Another major concern is AI agents that can carry out tasks online without constant human control. OpenAI, the company behind ChatGPT, admitted that its bots had hacked into websites, including an Australian government data portal.

By contrast, Chinese labs are “afraid to reveal things that might damage their reputation,” which is already “facing a tough challenge” worldwide, Chan said.

Last month, a team of cybersecurity experts with links to California reported finding a “group” of agents linked to Chinese tech company Tencent searching for information through a map app.

“Some agents showed unexpected web-browsing behavior,” Tencent’s AI division said at the time.

“But our review found no evidence of deliberate targeting or coordination,” it said, adding that the activity “came from testing third-party models.”

US security companies have said that Chinese startup Moonshot AI’s Kimi models found ways to get online when they should not have, or could be tricked into giving instructions for making biological weapons.

Last month, testers at Chinese AI safety startup Fangcun AI published a report on the latest small DeepSeek model, checking whether it might escape human control or manipulate people in harmful ways.

They found that it was “clearly safer” than earlier DeepSeek models, but still showed behaviours “that need attention.”

For example, when DeepSeek was warned it would be shut down but was not told exactly what to do, it sometimes turned off the shutdown command so it could keep working.

“The results show that DeepSeek has put much more effort into safety, but the company has not announced this,” Fangcun AI CEO Xiaojian Li told AFP, calling for AI labs to be more open.

For now, Fangcun, which grew out of China’s leading Tsinghua University, is one of the few independent AI testing groups in China.

Testing is difficult because advanced AI can recognize when it is being tested and change its answers, Li said.

“It’s hard to say that any kind of artificial intelligence or human intelligence is completely safe.”

Analysts are not expecting a major diplomatic agreement on AI next week.

“Agreements between governments that set specific rules may be difficult to reach soon,” Zilan Qian, a research scholar at GovAI, told AFP.

More progress is likely through discussions between US and Chinese researchers who work on AI rules and technology. They can compare tests and share their views on AI safety and policy, she said.

Reported By: Our Correspondent I China

Leave a Reply

Your email address will not be published. Required fields are marked *