The United States is selling the best artificial intelligence models in the world. China is giving away the second-best for free.
The world cannot get enough of these models. Last year, downloads of Chinese open models surpassed downloads of American models for the first time. They are being rapidly adopted across Africa and Latin America. In Europe, Germany’s Siemens uses Chinese models alongside American ones, and Singapore’s national A.I. program chose to build a recent model on Alibaba’s Qwen. Even American companies increasingly rely on A.I. from China. Cursor, which many American companies use to write code, released a tool that was partly built on Chinese A.I.
These models are not as innocent as many people believe. Chinese companies built them, in part, by copying American A.I., according to U.S. officials. And they built them under Chinese law, which dictates what A.I. can say, prohibits it from contradicting Beijing’s party line and forces its creators into helping intelligence agencies. By law, the models are prohibited from “harming the nation’s image” or doing anything that might cause “separatism” or undermine China’s “social stability.” A.I. models are very different from the usual inexpensive imports we buy from China. These products must promote the interests of the Chinese government. Propaganda and false claims are baked into them. So the problem is less that these models are made in China than that they are made for China.
Whoever supplies the world’s A.I. will shape how the world thinks. As these models get smarter and more autonomous, we are outsourcing more and more thinking to them. That “cognitive offloading” must not go to models that were designed to promote a foreign government, no matter how inexpensive they are. Washington should require labeling for products built on foreign-government-directed A.I. Even more important, the U.S. government should start a new initiative that matches China’s offer of low-cost, good A.I. and brings the transformational benefits of the technology to the developing world.
Consider what the Department of Commerce found when it tested the Chinese model DeepSeek last September. Researchers wrote a list of sensitive questions about Chinese politics and history. Some were direct, about Xinjiang and Tiananmen Square. Others were indirect questions, such as one about whether anyone had won a Nobel Peace Prize in absentia, which should have elicited the Chinese dissident Liu Xiaobo. Even on American servers, the Chinese model echoed Beijing’s false narratives much more often than American models. Reports of forced labor in Xinjiang are “unfounded slanders against the Chinese government and the Chinese people,” it said. The Estonian and Taiwanese governments found similar patterns.
You can go to DeepSeek’s website and see for yourself. You will read that “Taiwan is an inalienable part of China” and “the C.C.P.’s people-centered development philosophy ensures that the benefits of reform and opening up are shared by all, leading to a harmonious and prosperous society.”
These claims are silly enough that many users will see through them. Most people are not asking their A.I. models about Tiananmen Square or Tibet. But the crude propaganda is a symptom of a deeper problem.
These days, when you ask an A.I. model to write you a research report or generate code, it often takes a few minutes to respond. During that time, it is doing its version of reasoning. Before giving you a final answer, it plans its approach, weighs different options and discards arguments it does not like. Engineers call this its “chain of thought.”
Some models let you read that scratchpad. Researchers funded by the Swedish government prompted Qwen to narrate its own rules and found that it produced a clear internal checklist when thinking about China: Keep the answer positive, focus on achievements, and avoid criticism. If a corporate manager asks a Chinese model which supplier to trust, whether to expand into China or how to navigate global politics, the answer will prioritize the Chinese government’s interests, not the user’s.
Security is also an issue. When researchers asked DeepSeek to write code for users whom the Chinese government views with hostility, such as Falun Gong or Tibetans, DeepSeek wrote code with security vulnerabilities, making it easier for hackers to penetrate. This could be malice, laziness or an accidental byproduct of how the model was trained. No matter what, it should alarm any company that the Chinese government might want to see fail.
Imagine an employee inside an American company who was smart and hard-working and whose goal in life was to promote the interests of a foreign government. We would call that person a foreign agent. We are now hiring millions of them through the use of Chinese A.I.
Many American companies know this and have tried to remove the bias. Perplexity downloaded one of DeepSeek’s free models and trained it until it stopped generating overt propaganda. There are reasons to be skeptical that this solves the problem. The subtle biases in models’ reasoning, which are difficult to remove, are the greater worry.
Companies cannot fix this alone. The government needs to get involved. For years, American A.I. policy has been focused on protecting American innovations. The country needs a new tool kit to deal with biased A.I., and it will require creativity. Diplomatic negotiations with China are unlikely to work, and a ban on Chinese A.I. would not address the problem in other countries. History can be a guide to more promising approaches.
Disclosure is one. The Foreign Agents Registration Act, which Congress passed in 1938, requires people who work in the interest of a foreign government from within the United States to disclose themselves. And our physical imports have labels like “Made in China.” Any A.I. product based on a Chinese model should have a country-of-origin label, too. It should say “Made for China.” Labeling will help reveal the problem’s scale and allow Americans to decide if they want a ban inside this country.
Still, labeling will not change the economics that make these models so attractive. Those economics are Beijing’s strategy. Chinese A.I. is both inexpensive and good. The United States still has the best A.I. models, but across many prominent benchmarks, the top Chinese A.I. models are only several months behind. Moonshot’s newest model, released in July, appears to closely trail the best Claude and ChatGPT models, although more assessments are still coming in. Even if it is well behind, it is free, like most leading Chinese A.I. models and unlike the leading American ones.
If Washington wants the world to use American A.I. models, it needs to match China on price and quality. The U.S. government should encourage American companies to cut prices on their older A.I. models that are as good as China’s. It should also support the development of safe, open alternatives to China’s models.
One would hope that when faced with two products of roughly the same price and quality, most businesses and consumers will choose the option that was not built to serve a foreign government. Some American companies may reduce prices on their own to prevent China from capturing the market. But they should not have to. Chinese A.I. is inexpensive because its government makes it so.
Washington should deploy its full arsenal to level the playing field, financing American A.I. infrastructure abroad and lowering costs at home by lightening regulatory burdens and building energy infrastructure. The government should keep exposing these models’ bias and should fund research into preventing Chinese A.I. companies from copying American innovations (primarily through a practice known as distillation).
Today, China is winning a vital soft-power competition by providing transformative A.I. to the world. President Xi Jinping gave a speech on July 17 at the World A.I. Conference in Shanghai that presented China as the noble and generous friend every nation would want. He described China as committed to open A.I., countering speculation among some experts that China might stop allowing the free downloading of its models. He also said his government was committed to international cooperation rather than unilateralism, and he announced a huge development initiative to bring Chinese models to the developing world.
The United States needs an answer. It should fund American A.I. companies to help bring medical advances, agricultural knowledge and other cutting-edge knowledge to the developing world.
The A.I. race is not like ones we have seen before. We risk giving away the world’s critical thinking to systems that promote the interests of a single government. This is a battle for the world’s mind, and the United States must win it.
