Filed under ai-training-data
Every brief tagged "ai-training-data".
- § 01 · DATA-ASSETS
Two Registrations, One Word: China's New Data-Asset Standards and the Line Between 登记 and 登记
On 2 July 2026 China issued two national standards for data as an asset — GB/T 47949-2026 (classification and codes) and GB/T 47950-2026 (registration guidance) — both effective 1 September 2026. They give data assets a fixed place in the asset-classification code system (block A0806020000, including a first-ever asset code for AI-training multimodal data measured in tokens) and a step-by-step model for putting data on an organization's own books. The trap for overseas counsel is the word 登记 (registration): these MOF/SAC standards register data as an asset internally, while the National Data Administration's Data Property Rights Registration Work Guide (Trial), finalized 1 July 2026, registers rights in data externally through a certificated institution. Same word, two regimes, two artifacts, two purposes. This DCC brief separates them, reads the two standards for what they require, and explains why the 入表 (balance-sheet entry) vs 确权 (rights confirmation) distinction keeps tripping up data-asset deals.
- § 02 · TRANSPORT
Five Grades of Data, One Reporting Spine: The Ministry of Transport's Data Security Measures
On June 18, 2026 the Ministry of Transport issued the Measures for Data Security Management in Transport (交科技规〔2026〕3号), effective July 1, 2026 — 41 articles that complete the sector build-out of the Data Security Law for highways, waterways and comprehensive transport. The full text reached the public record in July through an academic-society WeChat repost rather than the ministry's own site. DCC reads the Measures around four load-bearing features: a five-grade classification ladder that splits general data into Grades 3/2/1 and pulls Grade-3 general data into the hard transmission-protection net alongside important and core data; an annual risk-assessment duty that extends beyond important-data handlers to any processor holding personal information on 10 million or more people, dated the same day as the national Network Data Security Risk Assessment Measures but effective 50 days earlier; an AI clause requiring pre-deployment evaluation of corpora, training data and algorithm explainability, plus a default ban on training on entrusted data; and a single reporting spine that routes filings through provincial transport authorities to MOT, with a direct line for central transport SOEs. Storage follows the sector pattern: localization for transport-authority personal information and CIIO-collected data, MLPS Level 3 for important-data systems, Level 4 or CII protection for core data, and security-assessed cloud services only.
- § 03 · JUDICIAL
Datatang v. Yinmu — China's First Ruling on a Data-IP Registration Certificate, and Why Open-Sourced Data Is Still Protected
A consolidated case study of 数据堂诉隐木科技 (Datatang v. Yinmu) — the Beijing IP Court's June 2024 appeal ruling, widely called China's first case on the evidentiary effect of a data-IP registration certificate. The dispute: Datatang built voice datasets for AI training, open-sourced some under a license; Yinmu took and redistributed them in the same data-services market. DCC synthesizes four commentaries (the case report, a Tsinghua analysis, and two Shenzhen Data Exchange DEXC+ deep-dives) into the four holdings that matter for overseas counsel: (1) a data-IP registration certificate is prima facie evidence of property-type interests and lawful sourcing — but not an absolute property right (property-rights-statutism); (2) open-sourced data, though neither trade secret nor copyrightable compilation, is protectable under the Anti-Unfair Competition Law's general clause; (3) the protection hierarchy (compilation work → trade secret → AUCL Art. 2); and (4) whether the taker honored the open-source license is the hinge for 'improper conduct.'
- § 04 · TOKENS
Cold Water on 'Token Trading' — Wang Qinglan on the NDA's High-Quality Data Set Initiative
In March 2026, the National Data Administration released the *Implementation Plan for Promoting High-Quality Industry Data Set Construction (Draft for Public Consultation)*, which explores a 'token (词元) based value system' and 'token trading as a new transaction mode' for high-quality data sets. The Chinese AI policy community immediately heralded the move as 'revolutionizing data trading.' Wang Qinglan pours cold water: token is a measuring unit, not a magic transformer. AI tokens are not crypto tokens. The bottleneck in China's data-element market isn't measurement — it's supply, rights clarity, compliance cost, and data silos.