China News Service, Beijing, October 16th: At present, artificial intelligence technology continues to be deeply integrated with application scenarios and gradually embedded into the social operating mechanism. With the rapid evolution of artificial intelligence technology, potential risks are also increasing. Once artificial intelligence is abused or used for malicious purposes, its harm will go beyond the scope of pure technology. It may not only cause local damage in key areas, but may even evolve into a systemic global crisis.
1. Effective governance of artificial intelligence is a global proposition
Uncontrolled and unregulated use of artificial intelligence technology is dangerous, and effective governance of artificial intelligence has become a common issue in the international community. The United Nations has clearly pointed out that effective governance is the key to helping artificial intelligence benefit mankind. Many countries have issued governance plans to provide a regulatory framework for the development and deployment of artificial intelligence and reduce its potential risks. For example, the United States has implemented supervision through multiple executive orders and export control measures, the European Union has implemented the Artificial Intelligence Law, and South Korea's Artificial Intelligence Basic Law, which will take effect on January 22, 2026, clarifies that operators of high-performance artificial intelligence must bear additional security obligations.
2. Supervision of artificial intelligence technology stack focuses on models and agents
On the basis of comprehensive legislation, countries have begun to refine specific regulatory measures from the technology stack level in response to the hierarchical architecture of AI. In terms of intelligent agents, the United States has completed the review process on intelligent agent supervision. Many comments proposed strengthening the supervision of multi-agent agents, especially monitoring from the development, deployment and other stages, and implementing model-level control measures. Singapore also released the "Agent Artificial Intelligence Governance Demonstration Framework" in January 2026, which requires early assessment and definition of risks, giving end users responsibilities, and strengthening the management and control of multi-agent agents.
Regarding the supervision of large models, take the United States as an example. During Trump’s second term in the United States, the AI proliferation rules during the Biden administration were revoked. However, the current 4E091 artificial intelligence model parameters are still on the Ministry of Commerce’s control list in Part 774 of the Export Administration Regulations, and are still controlled technologies. In order to promote advanced artificial intelligence innovation and safety, Presidential Executive Order No. 14409 in June 2026 stipulates the supervision of the deployment of security frontier models, requiring companies to "voluntarily" provide models to the government and conduct "pre-emptive security assessments" before publicly releasing frontier models. In response to Anthropic's release of its latest models, Mythos5 and Fable5, despite government warnings, the U.S. Secretary of Commerce requested in the form of an export control "notification letter" to restrict the access rights of foreign citizens. Anthropic subsequently removed the model from the shelves, cut off access rights, and worked closely with the government to eliminate relevant risks within two weeks. Only then did the U.S. Department of Commerce lift the restrictions in stages and conditionally allow the above-mentioned models to be open to "trusted partners." Currently, OpenAI’s latest model GPT-5.6 is also being released in phases to a limited extent in accordance with the instructions of the US government.
3. The “China Plan” of hierarchical classification helps to strike a balance between innovation and security
China has added four major categories of technologies in the field of artificial intelligence, including large models, agents, computing clusters, and information processing, to the "China's Catalog of Export Prohibited and Restricted Technologies", marking a new step in China's artificial governance. This is a new measure to prevent technology proliferation and a new paradigm to ensure that high-tech products are not used to harm national security and international peace. Taking the large models added this time as an example, China has implemented scientific hierarchical and classified management. Large models of core data are included in the prohibited category, and large models of important data are listed in the restricted category, which reflects the combination of supervision and end-use. The definition of data refers to the "Data Security Law", which demonstrates the scientific nature, accuracy and uniformity of listing management.
China's governance measures profoundly demonstrate the responsibility and pattern of a responsible major country. Faced with the widespread application of artificial intelligence in modern military conflicts, China resolutely prevents the weaponization of technology, maintains global peace with practical actions, and contributes a very constructive "Chinese solution" to the healthy and sustainable development of artificial intelligence. It also embodies the principle of coordinating high-quality development and high-level security. The country is committed to building an open source ecosystem. This technology listing is intended to strengthen the bottom line of security and clearly does not restrict normal exchanges in the open source field. It will not affect normal business behaviors such as cross-border joint research and development, the construction of overseas computing centers for enterprises, and overseas tape-out, leaving enough space for the scientific, sound and orderly development of technology.
