China's TC260 passed on 25 recommended national standard plans, 13 on AI and data security

Key takeaways

• The notice is a transmittal of standard plans. The attached list has 25 projects, 20 new standards and 5 revisions.
• By our count 13 items concern AI, including a foundation-model safety testing method, AI deep synthesis, safety guardrails and agent data processing.
• The list shows approval-submission deadlines of 2027-09-30 for the revisions and 2027-12-31 for the new standards.
• Nothing in the notice sets a duty, a certification step or a penalty.

Is the TC260 notice a binding standard, and who issued the plans?

No. The notice is titled "关于下达25项网络安全推荐性国家标准计划的通知" (Notice on Issuing 25 Recommended National Standard Plans for Cybersecurity) and carries the number 网安秘字〔2026〕121号, dated 13 October 2026. It is addressed to TC260 working groups and to the organizations that lead each project. It says the recommended national standard plans were issued recently by the Standardization Administration of China, and that 25 of them, including a JSON cryptography usage syntax specification, fall under TC260.

The notice asks working groups to prepare progress plans, to press lead organizations to carry them out, and to "广泛征求意见" (solicit opinions widely) so that standard quality is assured. It states no duty on companies, no certification scheme and no penalty. The attached list gives the standard nature of every item as recommended. We read these as plans for recommended standards, and the notice does not say whether any will later be referenced in a rule.

Which of the 25 items concern AI?

Counting from the attached list, 13 of the 25 items concern AI or AI-related data handling. Eleven sit with working group WG9 and two with WG8. One quoted title from the list is "网络安全技术 人工智能基础模型安全测试方法" (Cybersecurity Technology: AI Foundation Model Safety Testing Method), and another is "数据安全技术 智能体数据处理安全要求" (Data Security Technology: Agent Data Processing Security Requirements). Titles below are our English renderings of the Chinese list.

Other items cover cryptography, public key infrastructure, penetration testing, vulnerability description, mobile terminal personal information, cloud computing and an intelligent computing cloud service assessment method. We did not count those as AI items. A reader who counts the intelligent computing cloud item or the large-personal-information-processor data centre item as AI-related would reach a different total.

What are the timelines and the lead organizations?

The attached list gives every item a plan date of 2026-09-30. Twenty items are new standards with a 15-month cycle and a date to submit for approval of 2027-12-31. Five items are revisions with a 12-month cycle and a date of 2027-09-30. All 13 AI items in our table are new standards on the 15-month cycle.

The list names a lead organization for each item. The agent data processing item lists Beijing Institute of Technology, the AI-generated content detection guide lists Alibaba Cloud Computing Co., Ltd., and the foundation-model testing method lists the China Information Technology Security Evaluation Center. The notice asks working groups to press lead organizations to carry the plans out, so these are the bodies to watch for calls for comments.

What should AI vendors, evaluators and counsel do now?

Treat the list as a schedule of planned projects and not as a compliance requirement.

• Map your products against the 13 AI items, especially agent data handling, guardrails, open-source model security and AIGC detection.
• If you run AI security evaluations, note the plan for basic capability requirements for AI security evaluation bodies, since the attached list names it as a project.
• Watch the TC260 site for draft texts, because the notice asks working groups to solicit opinions widely before approval.
• When advising on China market entry, describe these items to clients as scheduled plans with late-2027 deadlines, not as existing standards.

What we did not verify

What we opened: the TC260 notice page and its attached 5-page PDF list (Wang An Mi Zi [2026] No. 121), both on 25 October 2026.

What we did not open: any draft or final standard text, the Standardization Administration of China issuing document, the national standards platform, and any separate project proposal for an item. The English item names are our renderings of the Chinese titles. The AI count of 13 is our own classification of the list.

What we refuse to claim: that any item will become mandatory, that a standard will be issued on time, what an item will require, or how regulators or courts will use it.

Related briefings

• AI Regulation News hub
• Asia-Pacific AI regulation tracker
• TC260 WG9 drafting of five AI security standards
• TC260 sector guides on AI application security
• TC260 guide on AI agent deployment security
• TC260 national standard on AI security classification and grading