KEY POINTS
- Administration officials privately expect little progress on AI safety and espionage concerns at Thursday's Trump-Xi meeting.
- A planned AI incident hotline between Scott Bessent and He Lifeng, announced in September, is still only a concept.
- The US has also moved ahead with a rule limiting China's remote access to US-made AI chips in foreign data centers.
Officials in the Trump administration expect little progress on AI safety and espionage concerns when President Donald Trump meets Chinese leader Xi Jinping on Thursday, according to WIRED's Inner Loop newsletter. The officials spoke privately to the outlet.
As he left the White House, Trump did not say whether he would raise Chinese interference in AI or the recent Nvidia scandal with Xi. Officials told WIRED the approach reflects a wish to avoid antagonizing Beijing during the summit.
The scandal involves Megaspeed, a firm described as having ties to China, and allegations that one of its executives was recruited by Chinese intelligence. China has rejected the claims, calling them groundless and a smear. Officials say the fallout has made any AI negotiation between the two largest economies more difficult.
In September, the administration moved to create an AI incident hotline linking Treasury Secretary Scott Bessent and Chinese vice premier He Lifeng, meant for situations like this. WIRED reports the channel is still at the concept stage and does not work yet.
Secretary of State Marco Rubio told reporters last week that the two governments have held AI safety talks at a technical level since the hotline was announced. He conceded that the effort is still a work in progress.
Administration officials have discussed a framework that would set out when each side must notify the other about an incident. It would also include enforcement measures so that the US is not the only party issuing alerts, officials said.
One obstacle is that the two countries see different dangers. US officials and tech leaders have focused on models becoming autonomous and misaligned, operating beyond human oversight. Chinese officials are widely understood to worry more about advanced open-weight models being used by critics and dissidents to undermine the Communist Party. The Chinese government has also shown interest in using frontier models to gather intelligence at home and abroad, so it would be unlikely to disclose a model going rogue, officials said.
Mutual suspicion of espionage has deepened the distrust, officials told the newsletter, and China had already seen the US as trying to hold back its AI development before the Nvidia episode.
Even as Trump pursues friendlier economic ties with China, the administration announced this week that it will go ahead with a rule restricting China's ability to remotely access US-made AI chips in foreign data centers. WIRED reports the rule followed criticism that a China-linked firm obtained advanced chips, and it could also bar companies with Chinese corporate parents from doing business with Nvidia.