Taipei, Taiwan — In late January, China’s leader Xi Jinping told a meeting of officials from across the country that China was on the brink of a “major technological revolution of historic importance.”
Artificial intelligence, he said, is as transformative as the steam engine, electricity and the internet. But despite all its potential, China cannot allow the new technology to “get out of control,” Xi warned during a study session for Chinese Communist Party leaders, according to state media. China needs to act early and decisively, anticipating and preventing problems with prudence and caution, he said.
Also read:
Continues after advertising
Xi’s statements highlight a tension shaping the country’s technology industry. Chinese leadership has decided that AI will drive economic growth in the next decade. At the same time, it cannot allow new technology to destabilize Chinese society and the Communist Party’s control over it.
The result is that the government is pressuring Chinese AI companies to do two things at once: move fast so that China overtakes international rivals and is at the forefront of technological change, while complying with an increasingly complex set of rules.
When Zhipu AI, one of China’s most promising AI startups, filed to go public in Hong Kong last month, valuing the company at more than $6 billion, it warned investors about the substantial burden of complying with half a dozen or more AI-related regulations. Similar to OpenAI, Zhipu develops AI models and applications that use them, such as the ChatGLM chatbot.
Among specific compliance requirements, the rules required Zhipu to act as a gatekeeper to prevent the spread of information that the Chinese government considers illegal. Zhipu could not be assured that regulators would always find the company compliant, which “could expose us to significant legal, financial and operational consequences,” it said in a filing.
The internet was once seen as an existential threat to the ruling Communist Party, but Beijing eventually brought it under control through a system of censorship and heavy oversight of the country’s biggest internet companies. Artificial intelligence presents a similar dilemma: a transformative force that promises economic gains while also having the potential to weaken the party’s grip on power.
But China’s regulatory safeguards add another layer of restriction and put Chinese companies like Alibaba, DeepSeek and Zhipu in the difficult position of racing to develop AI systems as powerful as those of foreign rivals like OpenAI and Anthropic, while also keeping their products compliant with more rules than their global competitors.
Continues after advertising
“What OpenAI and Alibaba are legally required to do in terms of testing before launch is quite different,” said Scott Singer, a researcher at the Carnegie Endowment for International Peace.
California’s new AI law, for example, requires companies to plan for the possibility that an AI system could surpass human control or cause a large number of deaths. Chinese rules, in contrast, focus much more on information control and data protection, Singer said.
Since 2022, Chinese technology companies have been required to provide the government with details about the algorithms behind features such as infinite video scrolling on apps such as RedNote and the domestic version of TikTok, Douyin.
Continues after advertising
Companies need to tell the government how their apps work and keep authorities up to date as technology evolves. The greater influence regulators perceive an app to have on public opinion, the more attention they pay to it.
But the government has sought to avoid limiting innovation or companies’ ability to experiment with new technologies, said Jiang Tianjiao, an associate professor at Fudan University.
“China does not want strict legislation and rules, especially binding law standards, to harm companies’ incentives to innovate,” Jiang said.
Continues after advertising
After ChatGPT sparked a global chatbot craze, Chinese officials quickly announced draft rules for AI systems capable of answering questions, holding conversations, and creating images and videos.
These systems learn by absorbing large volumes of data, and early Chinese models, like those developed elsewhere, were trained with widely available open source models such as Meta’s Llama. These models include internet sources such as Reddit and Wikipedia, which contain information censored in China.
Initially, officials drafted a standard that required all sources of training data to comply with government information controls and reflect “core socialist values.” If it had gone into effect, this rule would have set Chinese technology companies back months in training new models without input from the broader internet.
Continues after advertising
When the final decree was released in July 2023, however, the requirements had changed. Companies were now required to ensure that the information produced by their AI systems complied with China’s information controls only if this production was accessible to the general public.
“Chinese regulators think that AI technology will be transformative and that therefore the state should direct how it will be transformative,” said Graham Webster, a professor focused on geopolitics and technology at Stanford.
But Chinese authorities need to consider “whether enforcement of laws and regulations could undermine entrepreneurship and business vitality,” he said.
In December, Beijing announced a new set of draft policies for AI services that offer “human-like” interaction, such as the popular companion chatbot Xingye, made by Chinese startup Minimax.
These new rules seek to prevent people from becoming dependent on companion chatbots. Companies should not have “design goals of replacing social interaction,” the rules state.
That draft set of rules contained several provisions that analysts say are technically difficult to comply with.
As with the initial rules on generative AI, China has again stipulated that companies must ensure that chatbots are trained exclusively with government-approved information.
Companies are also required to create emotional profiles of users so that they can intervene if they show signs of wanting to harm themselves.
As Chinese officials seek to balance the promise of artificial intelligence with risk aversion, the technology has become increasingly important to Beijing amid slowing economic growth.
Since Chinese startup DeepSeek launched its popular AI model in late 2024, followed by a series of high-performance launches from Alibaba and Zhipu, local governments across the country have rushed to proclaim that they are using AI in every part of their economies, from hospital wards to industrial production lines.
“Because economic growth is so important to the party’s legitimacy, it wants to ensure that AI rules are not so burdensome that they stifle growth and drive away investment,” said Carnegie’s Singer.
c.2026 The New York Times Company
