Licensing Chinese Court Judgment Data for Legal AI: The 2026 Guide

Jurisdiction coverage has become a competitive front in legal AI. Harvey raised $200 million at an $11 billion valuation in March 2026; Legora was reported a month later at a $5.6 billion valuation with annual recurring revenue past $100 million; and both announced Singapore and Tokyo offices this year. As the category pushes into Asia-Pacific, one question lands on every product roadmap: what is the plan for China?

China's courts have published the largest body of judicial decisions ever made public — roughly 143 million effective judgments on China Judgements Online through December 2023, according to figures cited in Supreme People's Court (SPC) press briefings. Yet as of mid-2026, the public coverage pages of the major Western research platforms do not list mainland Chinese primary case law. The corpus exists; licensed, machine-readable access is the bottleneck.

This guide is for the teams closing that gap: legal AI product managers, law firm knowledge managers, and AI lab data teams. It covers why the gap persists, how public supply has shifted since 2020, the commercial structures used to license Chinese case law, the key compliance questions, and the due diligence to run before signing.

Why China Is the Last Major Jurisdiction Gap in Legal AI

Start with what incumbents publicly offer. vLex — acquired by Clio for $1 billion in November 2025 — advertises 110 jurisdictions and more than 1 billion documents. As of mid-2026, however, its public China coverage page lists international collections and secondary commentary, and mainland China does not appear in its published list of primary law jurisdictions. LexisNexis offers a Chinese legal library (LNCHNL) described as roughly 3,000 selected documents in English translation, and Westlaw China's site was not accessible when we last checked. That is no criticism of any provider — it is simply the published state of coverage, and it means Chinese case law cannot be sourced from the usual aggregators.

The AI-native platforms show why the gap persists. In July 2025, Harvey integrated official free law sources from eight countries in a single release — Austria's RIS, Finland's Finlex, and India's Indian Kanoon among them. The playbook is clear: where a government publishes case law as open, bulk-accessible data, plug it in. China does not fit it — China Judgements Online is a search portal rather than an open bulk-data program, and the newest national judicial database is internal to the courts (details below).

Where free sources do not exist, platforms license. Legora signs per-jurisdiction content partnerships — Otto Schmidt in Germany, Djøf Forlag in Denmark — and Harvey announced a content partnership with LexisNexis in June 2025 on undisclosed terms. China is the largest jurisdiction where neither the free-source route nor an established licensing pipeline exists yet.

There is also a corpus-scarcity angle. Surveys of website content languages put English near 59.8 percent of the web and Chinese around 1.3 percent — down from 4.3 percent over 11 years. High-quality Chinese legal text is scarce relative to China's economic weight, which makes a structured judgment corpus valuable as both grounding and training data.

The Supply Picture: 143 Million Judgments, a Narrower Pipeline

China Judgements Online (CJO) had accumulated about 143 million effective judgments by December 2023 — the largest public collection of court decisions in the world — and has recorded more than 108.1 billion cumulative visits. The trend line matters more than the headline. Annual publication volumes, as cited in SPC press briefings and the SPC's March 2026 work report:

Two structural changes in early 2024 reshaped access. In January 2024, the new National Court Judgments Database went live for use inside the court system only — lawyers, researchers, and the public cannot search it. On February 27, 2024, the SPC opened the People's Court Case Library, a curated set of SPC-vetted reference cases that launched with 3,711 cases, growing to about 5,327 by January 2026 — a citation-quality reference library measured in thousands of cases, not a bulk corpus.

The practical read for data buyers: the historical corpus through 2023 is the strategic asset, and the ongoing public flow, while recovering, remains well below peak. For a deeper analysis of what is and is not retrievable in 2026, see The State of Chinese Case Law Data in 2026.

What a Licensable Chinese Court Judgments Database Looks Like

CJO publishes judgments as individual web documents — workable for looking up one case, unworkable for building retrieval systems, extraction pipelines, or training sets. A licensing conversation is really about the engineering on top of the public record: collection at scale, deduplication, normalization, and parsing into consistent machine-readable fields.

SinoVerdict's corpus covers 170M+ Chinese court judgments as a structured snapshot through 2023, with incremental updates available on request. Each record pairs full judgment text with structured metadata — court, region, date, and case-type information — so teams can filter, sample, and benchmark by slice instead of ingesting blind. Delivery matches the three ways teams consume legal data:

• REST API — query-time retrieval for RAG and agent workflows
• Bulk dataset — full or scoped exports for indexing, analytics, and model training
• MCP server — a Model Context Protocol endpoint that lets AI assistants and agents search the corpus natively

Schema design determines how useful any of this is in production. We document record structure, field semantics, and query patterns in our companion article on the Chinese case law API structure.

Three Commercial Structures for Chinese Legal Data Licensing

There is no single market-standard contract for licensing Chinese court judgments, but most negotiations converge on one of three structures, often in sequence.

The per-jurisdiction license is already proven in legal AI — it is how publishers such as Otto Schmidt and Djøf Forlag partner with Legora in Germany and Denmark. Expect the longest negotiation over the definition of attributable revenue, audit rights, and treatment of indexed data at termination.

API subscriptions trade control for speed: fast integration and vendor-managed updates, in exchange for attention to rate limits, caching and output-retention rights, and uptime commitments. In every structure, model-training rights and end-user display rights are distinct bundles, typically priced separately.

Pilots are where serious buyers start. A useful pilot fixes its benchmarks up front — retrieval relevance on your own query set, extraction accuracy on sampled fields, coverage checks by year and court level — and credits pilot fees against a full license. SinoVerdict runs evaluation pilots through its trial API program.

Compliance Overview: Anonymization, Transfer Pathways, and DSL Article 36

This section is general information for procurement planning, not legal advice. Engage qualified counsel on your specific transaction and use case.

Four topics come up in nearly every negotiation over Chinese legal data:

• Anonymization vs. de-identification. Under China's Personal Information Protection Law (PIPL), de-identified data remains personal information, while genuinely anonymized data falls outside the law's scope. Published judgments occupy a specific position: under SPC publication rules, party names are retained, while identifiers such as national ID numbers and home addresses are redacted before a judgment goes online. How that status interacts with PIPL obligations in downstream processing is a question for counsel.
• Cross-border transfer pathways. Chinese law provides three mechanisms for moving regulated data out of the mainland: a government security assessment, standard contractual clauses, or certification. Which, if any, applies depends on the data, volumes, and parties involved.
• DSL Article 36 and foreign discovery. Article 36 of the Data Security Law prohibits providing data stored in China to foreign judicial or law enforcement authorities without approval. License agreements for Chinese-origin data therefore commonly exclude use for foreign discovery or law-enforcement production. If your product serves litigation teams, surface this restriction early.
• Domestic deployment. Some buyers process the data inside China — an in-country deployment of the index or pipeline, with only derived outputs leaving. Whether that changes the regulatory analysis is for counsel, but it is an architecture worth pricing.

A capable vendor should state its position on each of these in writing — exactly what the diligence list below is designed to test.

Eight Due Diligence Questions Before You License Chinese Case Law

Whichever vendor you evaluate — SinoVerdict included — put these eight questions on the table:

1. What are the exact document count and date coverage? Ask for a coverage report by year, court level, and region, and reconcile it with the public CJO timeline above.
2. What is the provenance of the corpus? Confirm it derives from publicly published judgments and ask how collection is documented.
3. How is deduplication handled? One dispute can generate multiple documents across procedural stages; ask how duplicates and related documents are linked.
4. Which structured fields are guaranteed, and at what completeness? Field-level fill rates by year and court level matter more than a schema diagram.
5. Which rights are in the bundle? Internal evaluation, model training, end-user retrieval and display, and redistribution are distinct rights — get each in writing.
6. What happens when source documents are withdrawn? Ask for the vendor's policy if a judgment later leaves public access and how removals propagate to your copy.
7. What compliance representations does the contract make, and what does it exclude? Expect language on redaction status and the cross-border mechanism relied on, plus an express exclusion of foreign judicial or law-enforcement use consistent with DSL Article 36.
8. What are the update terms? Is the dataset a static snapshot, and on what cadence and at what price do incremental updates arrive?

A vendor that can answer all eight in writing is ready for an enterprise contract; hesitation on any of them is a finding in itself.