Skip to content
DCC · DATA COMPLIANCE CHINA China data law, for overseas counsel.
§ 120 · CAC

What Is New in the CAC's Filing FAQs: Same Model at Different Parameter Sizes Is One Filing, and Algorithm Filing Counts Algorithms, Not Products

数据何规's same-day note on the guidance posted with the new filing portal at beian.cac.gov.cn, translated in full: four points the author believes the CAC had not stated at national level before

A full translation. On November 9, 2026, the day the Cyberspace Administration of China opened the Internet New Technology and New Application Information Service Filing System at beian.cac.gov.cn, 数据何规 pulled four points from the FAQ documents on the portal's home page that it believes are stated at national CAC level for the first time. For generative AI: an application or function that directly calls an already-filed large model through an API may be registered locally, which the author reads as 'large-model registration goes to the provincial cyberspace administration'; and large language models with the same training corpus, algorithm and security safeguards that differ only in parameter size may be regarded as the same model and need not be filed again. For algorithm filing: the algorithm is the basic unit, so one algorithm used in several products is filed once and linked to each product; and review follows the order of submission, so an application resubmitted after modification queues again.

Editor’s Note — DCC.

A short practitioner note, translated in full. The author quotes four answers from the CAC’s new FAQ documents and adds a one-line reading of each. The complete FAQs are on DCC’s law page, CAC Unified Filing System Notice and FAQs (2026), and DCC’s account of the portal is in beian.cac.gov.cn Goes Live. 备案 is “filing”; 登记 is “registration”. The claim that these points are new at national level is the author’s; DCC adds two cautions at the end.

Today, the Cyberspace Administration of China brought its filing system for new internet technologies and new applications online.

In order to further raise the level of standardization of filing work for new technologies and new applications, and to make it convenient for enterprises and the public to submit and look up filing information for new technologies and new applications online, the Cyberspace Administration of China has built the “Internet New Technology and New Application Information Service Filing System”, which can handle filings for generative artificial intelligence, algorithms, deep synthesis, blockchain and other new technologies and new applications in an integrated manner. The system formally went into operation on November 9, 2026, at https://beian.cac.gov.cn. The former “Internet Information Service Algorithm Filing System” and “Blockchain Information Service Filing Management System” ceased to be used at the same time.

For any questions arising in the use of the system, please call 010-82990521. Calls are answered from 9:00 to 17:00 on working days.

— 网信中国, Notice on the Launch of the Internet New Technology and New Application Information Service Filing System

There are some guides on the home page of the website. Some of the information, I feel, is being stated at the level of the Cyberspace Administration of China for the first time. I excerpt it here for readers’ reference.

Generative AI

2. In what circumstances is generative AI service registration available?

A generative AI application or function that directly calls the capability of an already-filed large model through an API or by other means may, on the premise that it satisfies the relevant laws and regulations of the State, complete a generative AI service registration with the local authority.

3. In what circumstances is there no need to file a generative AI service again?

Large language models whose training corpus, algorithm used and security safeguards are the same and which differ only in parameter size may be regarded as the same model, and need not be filed again.

— Compiled Answers to Frequently Asked Questions on the Filing of Generative AI Services (Including Anthropomorphic Interaction)

First, it stresses once again that large-model registration goes to the provincial cyberspace administration.

Second, it makes clear for the first time that “large language models whose training corpus, algorithm used and security safeguards are the same and which differ only in parameter size may be regarded as the same model, and need not be filed again”. This is very important.

Algorithm filing

3. Do multiple products need to be filed multiple times?

Internet information service algorithm filing takes the algorithm as its basic unit.

Where multiple products use the same algorithm, there is no need to file repeatedly; the algorithm need only be linked to the multiple products.

Where multiple products use multiple algorithms, the different algorithms must be filed separately, and each algorithm linked to the corresponding product.

4. In what order are internet information service algorithm filings reviewed?

At present, review follows the order in which filing applications are submitted: first submitted, first reviewed. For an application resubmitted after modification, the time of resubmission governs.

— Compiled Answers to Frequently Asked Questions on Internet Information Service Algorithm Filing

First, algorithm filing works on the dimension of the algorithm, not the dimension of the product.

Second, a filing resubmitted after modification has to queue again.


DCC’s cautions

The notes below are DCC’s, not the author’s.

  • “Provincial” is the author’s word; the FAQ says “local” (属地). The answer on registration says it may be done 在属地, and the answer on how to apply says materials go to the 属地网信办. The portal’s consultation column lists provincial cyberspace administrations’ contact details, which supports the author’s reading in practice, but the FAQ text does not name the level.
  • “Same model” has three conditions, and each is a question of fact. The relief from re-filing applies where training corpus, algorithm and security safeguards are all the same and only parameter size differs. A distilled or quantized variant trained on a different corpus, or one shipped with different safety measures, is not covered by the sentence as written. The FAQ addresses “large language models” only.
  • Registration is described as available, not as required. The FAQ says an API-based application “may” (支持) register locally. It does not say that an unregistered application is in breach. For the enforcement side, where the CAC warned an “API relay station” that had not carried out the required security assessment, see case 10 in CAC’s 15 September 2026 Enforcement Typical Cases.

— 数据何规, 算法/大模型备案官方问答的增量信息 (The Incremental Information in the Official Q&A on Algorithm and Large-Model Filing), November 9, 2026. Full translation by DCC; a screenshot of the portal’s guide list is omitted. Original (Chinese).

— Not legal advice.

— Not legal advice.


§ SUBSCRIBE

The Monday brief.

One short email every Monday. New briefs on Chinese data-compliance rules from the previous week, with the source law cited.

Opt-in only. Unsubscribe anytime by replying "unsubscribe" to any issue.

SUPPORT DCC

Keep the publication free to read. Suggested support is $19.99, or choose your own amount.

Support →