At the Shanghai World Artificial Intelligence Conference, Chinese Communist Party leader Xi Jinping advocated for the “open source and open collaboration” of artificial intelligence. However, Chinese developers have informed reporters that the authorities maintain internal network censorship and political scrutiny, with some technical articles and overseas websites frequently encountering “404” errors. To access overseas models, data, and technical information, developers are forced to use circumvention tools.
Xi Jinping, during the 2026 World Artificial Intelligence Conference and the High-Level Meeting on Global Governance of Artificial Intelligence held in Shanghai, announced plans to provide 5,000 AI study slots to developing countries over the next five years. He also proposed the establishment of an International AI Application Cooperation Center for countries such as ASEAN, the Arab League, African Union, the Community of Latin American and Caribbean States, the Shanghai Cooperation Organization, and BRICS nations.
Chinese AI developer Wu Zhennong told reporters that while authorities promote the openness of AI, they still block information internally, denying netizens access to overseas information. Wu stated, “When we are developing, we often face difficulties in accessing overseas models, datasets, and technical discussion forums. Downloading models requires changing mirrors, altering ports, and sometimes using proxy tools, which is a hassle. Official policies demand that we keep up with international technology, yet ordinary netizens are kept within the constraints of the Great Firewall. The so-called ‘open source’ is just a deception aimed at countries in Asia, Africa, and Latin America.”
Wu Zhennong believes it is ironic that a regime relying on internet censorship to maintain control promotes the openness of artificial intelligence. He remarked, “It sounds ironic to me, and netizens also see this as a self-inflicted insult.”
Reporters discovered that as early as July 13, 2023, the Chinese National Internet Information Office and seven other departments issued the “Interim Measures for the Management of Generative AI Services,” effective from August 15 of the same year. The measures stipulate that services providing generated content to the Chinese public must adhere to the “core socialist values” and avoid producing content deemed subversive, harmful to national security, damaging to the country’s image, or disruptive to societal stability.
In the first half of this year, netizens testing various Chinese AI tools reported that when dealing with sensitive issues such as the June 4th incident, the Blank Movement, human rights in Xinjiang, internal power struggles within the Communist Party, or Xi Jinping himself, the models often declined to respond or output official Communist Party rhetoric.
Chinese internet technology scholar Gu Liming (pseudonym) told reporters that the Chinese Communist Party’s control over AI tools is primarily aimed at maintaining the legitimacy of its rule and functions more as a propaganda tool. He remarked, “Our developed AI models automatically avoid political and critical content. The Party requires models to comply with political security, and companies must remove sensitive content, while developers cannot openly discuss errors. The ‘big boss’ talks about openness to absorb global technology, but this is not open source; it is for greater ‘control’ domestically. Openness has been replaced by struggle; where is the openness?”
Hugging Face is a global platform for sharing AI models and datasets, and several Chinese models depend on this platform for dissemination and testing. However, when Chinese developers use such overseas platforms, they often encounter issues like connectivity failures, slow downloads, and unstable service. Chinese technology websites have long provided guidance on accessing domestic mirrors, modifying download endpoints, and using proxies to visit.
AI engineer Mr. Lu told reporters that after the Great Firewall cut off access to many overseas websites, Chinese tech personnel often need to bypass the official network barriers to achieve the required AI innovation.
Regarding the development of Chinese AI technology, he said, “Recently, China’s Moonshot AI claims to be technologically leading globally, but it still relies on US technology as the basis, although it also has its innovations. Yet, I believe that before long, you will find that such technology still falls short of the US. Every time the mainland announces a new invention over the past few decades, it is soon debunked as old technology by foreign countries. I think this time will be no exception.”
In recent years, the US has restricted China’s access to advanced AI chips and semiconductor manufacturing technology. Beijing is accelerating the development of domestic chips, frameworks, and large models, while attracting overseas developers to enter the technical system established by Chinese companies by partially opening up model weights.
Despite this, existing Chinese regulations do not provide clear and stable cross-border internet channels for ordinary developers. Developers who use proxy tools to access overseas code repositories, paper websites, and technical forums may face interrogations or penalties, even if it is for work or research purposes. Mr. Lu disclosed, “They (the CCP) encourage AI innovation but have not eliminated the legal risks for technical personnel to access international data. Currently, some engineers dare not leave the country, and others are prohibited from doing so.”