On 5 January 2024 the National Data Administration and sixteen other departments published the three-year “Data Elements ×” Action Plan (2024–2026). The plan does not stop at principles. It lists twelve sectors — industrial manufacturing, agriculture, commerce and trade, transport, financial services, scientific innovation, culture and tourism, healthcare, emergency management, meteorological services, urban governance and green development — and assigns concrete data-fusion scenarios to each.
What the plan actually changes
Earlier discussion of data as a production factor centred on ownership, circulation and benefit-sharing. This plan moves the framework into sectoral scenarios, which means data supply now has an organisational form and a basis for assessment. Public-data authorisation, sector datasets, data exchanges and trusted data spaces become things that can be applied for and inspected.
For AI companies the consequence is direct: the acquisition of training corpora is being institutionalised. Scraping public sources, relying on customer authorisation or buying third-party datasets will increasingly require an explanation of provenance, scope of licence and limits on use.
Three points that matter for AI companies
- Corpus supply. Sector datasets and authorised public data create new lawful sources, but authorisations usually carry purpose, duration and non-transfer restrictions. They cannot be assumed to permit resale or open sharing.
- Circulation infrastructure. Trusted data spaces and data exchanges address verifiable transfer records, which aligns with regulatory expectations on provenance.
- Data security. Classification, identification of important data and outbound transfer rules continue to apply in parallel. The more active the circulation, the more specific the explanation of source and destination.
Three things to do now
- Map provenance and the licence chain. Separate public scraping, customer-provided data, third-party purchases and self-generated data, and confirm for each whether the licence covers training, fine-tuning and external service.
- Separate internal use from external provision. Training a model internally and exporting a dataset are governed differently; the latter more readily triggers licensing and outbound review.
- Build compliance into the product cycle. Provenance statements, licence evidence and processing records should exist before launch, not be assembled when a filing or inspection arrives.
For a provenance and licensing review, see our data compliance service and IP protection service. Where data is provided externally or across borders, start with the business licensing analysis. Definitions are collected in the glossary.
This article is based on the Data Elements × Action Plan (2024–2026) published by the National Data Administration and other departments on 5 January 2024, and on public information. It is for reference only and does not constitute legal advice.