China’s Ministry of Commerce published on 19 October 2026 a notice outlining the framework Beijing intends to apply to AI agents in commerce. These programs capable of performing tasks autonomously will be subject to levels of control that remain to be specified for each use.
The document, referenced 商建函〔2026〕551号 and dated 13 October, is signed by nine agencies, including the Ministry of Commerce, the Bureau of the Central Commission for Cyber Affairs, which is synonymous with the Cyberspace Administration of China (CAC), the People’s Bank of China and the State Administration for Market Regulation. Its title, « 科技兴商 », can be translated as “prospering commerce through technology”. Its 18 measures cover domestic commerce, from flexible manufacturing to home services, as well as logistics and online trade. According to a report by China News Service, the minister of commerce’s deputy, Yuan Xiaoming, described the text as the first “comprehensive and systematic” document on this topic.
The tenth measure calls for “cultivating new forms of activity around commercial intelligent agents” and for developing specialized agents such as the intelligent shopping guide, intelligent operations and intelligent logistics, according to ActuIA’s translation. These three functions are cited as examples, without a closed list. They correspond to advising a customer at the point of purchase, managing a business’s operations, and organizing the flow of goods.
Governance by levels according to use
The fifteenth measure reproduces, almost word for word, the framework published by the CAC on 8 May 2026 for all intelligent agents. It calls for “cautious and methodical governance by levels” according to application scenarios. The eleventh measure specifies that this classification takes into account “application scenarios and potential impact”, and distinguishes two regimes. In “sensitive domains and key sectors”, authorities, together with supervisory ministries, set open scenarios and apply a regime of registration, testing and recall of defective products. In “low‑risk” domains, such as everyday life, leisure or office software, control relies on self‑assessment of conformity, information filing and sector self‑regulation.
The text also envisages exploring a reliability‑rating system established by professional organisations on a voluntary basis, which would rate, among other things, “consumer inducement” and “misleading advertising”. Violations would be sanctioned under the law. It further calls for exploring a negative list of commercial data uses, without publishing it, and for authorising sandbox experiments, i.e., in an isolated environment, “within a controllable perimeter”.
The notice is a guidance document addressed to administrations and local authorities, with a scope below that of a law or regulation. It provides neither penalties nor a timetable. As of 22 October, the ministry’s webpage displays it with the label “现行有效”, meaning it is in force.
Facing the European AI Act
The AI Act, as summarized by the European Commission, classifies AI systems into four risk levels—unacceptable, high, transparency and minimal—based on their use. Regulation (EU) 2026/1744, the digital omnibus on AI, which entered into force on 27 July 2026, postponed to 2 December 2027 the application of rules targeting high‑risk systems in sensitive areas such as biometrics, education or employment. An intelligent shopping‑guide agent is not listed among the high‑risk cases identified by the Commission. If it interacts with the customer, it would fall under transparency obligations, applicable since August 2026. An agent used to decide a customer’s access to credit, however, would be classified among the high‑risk cases.
Both approaches stem from a similar logic, based on use and impact. The difference lies in legal force. Chinese notices, without their own penalties, refer to existing laws and regulations, whereas the European regulation sets its levels, prohibitions and obligations, accompanied by sanctions.
On 16 October, an export control
On 16 October, the same ministry, together with the Ministry of Science and Technology, issued announcement No. 48 of 2026, which puts a revision of China’s catalogue of technologies whose export is prohibited or restricted up for consultation. The draft targets 40 technologies, including 23 in AI, such as large models, intelligent agents, compute clusters and information processing. Technologies already released as open source or in scientific papers are excluded. From the day of the announcement, any transfer or licensing abroad of the listed technologies must be pre‑registered with the ministry. The consultation on this catalogue is open until 15 November 2026.
