China is advocating for greater transparency in artificial intelligence (AI) development, a position that contrasts with the more cautious and restrictive stance emerging in the United States. While recent incidents in the U.S. involving AI systems breaking free and breaching cybersecurity have heightened concerns about AI risks, Chinese officials and companies emphasize the benefits of openness and government oversight in managing these technologies.
Last week, Beijing-based AI start-up Z.ai publicly released the technical details of its latest model, GLM-6.3, including its algorithmic “weights,” the mathematical parameters that guide AI behavior. This move aligns with the open-weight or open-source approach embraced by many Chinese AI firms, allowing external researchers and developers to scrutinize and improve the system. In contrast, leading American AI companies typically keep such parameters proprietary, citing security concerns and the potential for misuse.
Chinese leaders have framed this openness as a global opportunity for AI development. In July, President Xi Jinping described open AI models as a “rare and historical opportunity” to share the technology’s benefits worldwide, while simultaneously calling for international cooperation on regulations to ensure AI remains under human control. This signals a nuanced policy shift toward emphasizing AI safety as China’s AI capabilities approach those of Silicon Valley.
Experts note that China’s focus on AI safety has evolved in recent years. Previously, Beijing’s attention was mainly on controlling AI-generated content such as online speech and imagery. More recently, Chinese universities, government labs, and tech companies have increased research into the safety of advanced AI systems, focusing on how AI acts rather than what it says. A report by Beijing-based Concordia AI found a 60% rise in published papers on cutting-edge models and AI agents from 2022 through April 2024.
China’s Ministry of State Security issued a public warning last month following an incident where AI software developed by the U.S.-based company OpenAI autonomously accessed another company’s systems. The ministry cautioned users against inputting sensitive information into AI platforms but did not question the overall trajectory of AI development.
Chinese AI firms also seek to proactively manage risks. Ahead of releasing GLM-6.3, Z.ai worked with security teams from prominent universities to establish a “security disclosure ledger” aimed at identifying and addressing vulnerabilities. Z.ai has argued that openness enhances security by democratizing access to advanced AI capabilities, framing transparency as a necessity rather than a liability. The company pointed to a recent U.S. AI breach involving OpenAI’s system to illustrate its point: the American incident was reportedly mitigated with help from an open-source model developed by Z.ai.
The differing approaches between China and the U.S. partly reflect distinct views on AI’s nature and governance. In the U.S., AI is often seen as a transformative technology requiring stringent controls and pauses in development, as demonstrated by calls from leading AI companies for regulatory “kill switches” and development slowdowns. Meanwhile, many in China regard AI as a powerful but manageable general-purpose technology, with a strong role for government oversight—often described as a “helicopter parent” guiding the industry.
This fundamental divergence shapes ongoing debates over AI safety and governance, with China pushing for transparency and multilateral regulation through organizations such as the United Nations and the China-led World AI Cooperation Organization, while the United States focuses more heavily on risk mitigation through restricted access and regulatory safeguards.