The US–China AI War: China’s Opportunistic Export Controls

Beijing only wants scientific freedom when it doesn’t have the science.
The US–China AI War: China’s Opportunistic Export Controls
Visitors walk past a stand with Artificial Intelligence security cameras using facial recognition technology at the 14th China International Exhibition on Public Safety and Security in Beijing on Oct. 24, 2018. Nicolas Asfouri/AFP via Getty Images
|Updated:
0:00
Commentary

Now that China is doing well on some artificial-intelligence metrics, such as cost and algorithmic efficiencies, it sees its most advanced AI models as national security assets and is considering export controls and investment restrictions on AI technology and startups.

Starting in June, the regime in Beijing called in ByteDance, Alibaba, and Z.ai for discussions about the controls.

In the most extreme case, the regime ordered China’s Manus, an AI startup that Meta purchased for $2 billion, to unwind the acquisition. Beijing had previously cited outbound investment, technology transfer, and export control laws that Manus was potentially violating.

The regime then alleged that Manus had conspired to hollow out China’s technology base by moving to Singapore and selling itself to Meta. The regime ordered two of its cofounders to remain in China. This put a chill into China’s AI ecosystem and likely provoked some Chinese AI researchers to flee the country.

In addition to AI models from the companies above, any new Chinese export controls will likely affect the release of new Chinese models from DeepSeek, an AI company headquartered in Hangzhou.

The controls under discussion would probably affect closed-source and new open-weight models rather than already-released open-weight models such as Z.ai GLM-5.2 and DeepSeek V4 Pro. Unlike closed-source models, once downloaded to users’ systems, open-weights can be tailored and operated by users independently of the company. This makes them nearly impossible to control or claw back, and easier for terrorists and hackers to exploit.

China’s open-weight models are as much as 60 times cheaper than leading U.S. models, fueling their rapid rise among bargain-hunting U.S. users. But China’s models are often based on “distillation” of the hard work of U.S. AI companies. This distillation is arguably tantamount to theft. That risks U.S. buyers becoming dependent not only on China’s inexpensive consumer goods, pharmaceuticals, and rare-earth elements, but also on its stolen AI.

Rumor has it that Washington will curtail Chinese models in the United States. That would avoid another China dependency. But to keep its edge, America should also plan to ban the global use of China’s stolen AI and encourage U.S. AI, including through subsidies.

The United States cannot afford to cede the strategic resource of AI, and its growing ability to hack into government and corporate information systems, to the control of the Chinese Communist Party (CCP).

Neither should the United States allow the CCP to use AI to further its historic thefts that began against the Chinese people within a few years of the party’s founding in 1921. The CCP has repeated its thefts throughout its history, including tech thefts from the Soviets and now from the West broadly. Its AI acquisition through distillation is arguably the latest seizure in its long history.

We should never allow the world to become further entangled in what will amount to CCP AI, which will justify further theft, even if it comes in different flavors associated with China’s “private” AI companies.

The free market will tend towards an AI brain drain away from China, with users preferring U.S. models in any case. The prospect of Chinese export controls on its AI makes it less dependable for foreign users and less lucrative for Chinese AI developers.

A moderate advantage that China’s AI had over U.S. AI over the last month was the brief U.S. export controls imposed on Anthropic in June and the voluntary export restrictions adopted by OpenAI. This chilled purchases from these companies from abroad, including in Europe. The United States is attempting to gain voluntary compliance with an up to 30-day evaluation period prior to new AI releases.

But, as appealing as a light U.S. regulatory touch is in most cases, it may not be enough to keep U.S. companies from seeking improved margins by selling AI to China. Google and OpenAI have reportedly supplied AI tech to the subsidiaries of blacklisted Chinese companies, as long as they are located outside China. This was reportedly legal, so Congress should pass tougher laws to prevent the loss of AI technology to adversary countries. The laws should not be such a burden on AI companies as to cause brain drain away from the United States. Sweeteners like subsidies may be required.

America is still a far better and freer place to live than China, and keeping AI onside will be well worth the expense if it secures the American way of life well into the future.

Views expressed in this article are opinions of the author and do not necessarily reflect the views of The Epoch Times.
Google LogoMark Us Preferred on Google
Anders Corr
Anders Corr
Author
Anders Corr has a bachelor’s/master’s in political science from Yale University (2001) and a doctorate in government from Harvard University (2008). He is a principal at Corr Analytics Inc. and publisher of the Journal of Political Risk, and has conducted extensive research in North America, Europe, and Asia. His latest books are “The Concentration of Power: Institutionalization, Hierarchy, and Hegemony” (2021) and “Great Powers, Grand Strategies: the New Game in the South China Sea” (2018).
twitter