+34
The hottest topic in the global artificial intelligence (AI) industry lately is the debate within the US government over regulating Chinese AI models. Following the release last month of the AI mod...
Reuters Yonhap News The hottest topic in the global artificial intelligence industry lately is the debate within the US government over regulating Chinese AI models.
Following the release last month of the AI model ‘Kimi K3’ by the Chinese startup Moonshot, which caused a stir by showing performance close to leading US models, calls within the US government to regulate China-made open-source AI have gained strength.
By contrast, US big tech firms that supply hardware such as AI accelerators and cloud services to open-source models oppose the move, saying regulation would undermine competitiveness. With AI expected to be a core agenda item at the US·China summit scheduled for next month, moves have emerged to tighten export controls to block China from accessing advanced US AI chips ‘remotely’ via the cloud and other means.
If US controls on cloud services were to materialize, some expect Washington could ask allies such as South Korea to join in the export-control effort.
CNBC, a US business outlet, reported on the 19th that the US government, which has restricted exports of the most advanced Nvidia chips to China, is seeking to plug loopholes in export controls by blocking access by China to computing resources through cloud providers. In practice, the cloud has been regarded as a blind spot in export controls.
Questions have been raised that Chinese companies may be securing advanced chips through data centers or cloud services in the United States and third countries, using them to train AI models or for commercial AI. The White House even claimed that, right after the launch of Kimi K3, model training was conducted by accessing the advanced GB300 chip from Nvidia located in Thailand.
According to the Carnegie Endowment for International Peace, at least 11 entities linked to the Chinese government attempted to access US technologies subject to controls from third-country clouds. In the US Congress, the ‘Remote Access Security Act ’ is also under discussion to regulate this.
After passing the House in January, a bipartisan version was introduced in the Senate and is awaiting a vote. The bill would prohibit remote access, including via cloud services, to US technologies subject to export controls.
The intent is to go beyond preventing China from physically obtaining equipment or chips and to cut off remote connections to advanced hardware·software. If the scope of export controls expands to access through the cloud, major US big-tech cloud service providers such as Amazon, Microsoft, and Google would be directly affected.
Implementation costs are also expected to rise, such as having to prove that service purchasers and end users have no ties to countries of concern including China. Some therefore argue that this could actually weaken the market dominance of US companies, as data centers in third countries such as Southeast Asia may turn to Chinese chips instead of US-made ones.
It cannot be ruled out that the United States will ask allies, including South Korea, to prepare control measures at a similar level. A diplomatic source familiar with the matter said, “A consistent US position is that participation by allies and partners is essential to increase the effectiveness of export controls on China.
” In a phone interview, Choi Byung-ho, a professor at the AI Research Institute of Korea University, said, “If the United States moves to regulate Chinese AI, the use of Chinese models by Korean companies could be affected depending on the scope,” while also saying, “Given that Chinese open-weight AI models are widely used around the world, a comprehensive ban will not be easy. ” 한글기사 원본
Alarmed By Chinese AI Surge Moves To Extend Export Controls To Cloud Services