China's IP office built a national AI system for patent search, translation and document comparison, live since January 2023 and expanded in 2025 with large-model search and legal-assistant tools — while officially barring examiners from using AI output as an examination opinion.
Details
Maturity
Established
Promoter
China National Intellectual Property Administration (CNIPA)
Period
2022-2025
Keywords
intellectual property, patent examination, government digitalisation, AI-assisted search
Context
China's National Intellectual Property Administration (CNIPA) began building a patent intelligent examination and search system in 2022, officially launched in January 2023. The system provides online translation, graphic recognition and intelligent document comparison, aimed at reducing examiners' repetitive work. By July 2025, CNIPA had added a large-model search tool, an AI academic assistant and an AI legal assistant offering natural-language queries over case law. CNIPA has explicitly stated that AI-generated results 'cannot be directly used as examination opinions' — examiners must reach conclusions independently under the Patent Law.
Results
At a June 2025 press conference, CNIPA officials cited a 'significant improvement in the reference detection rate' from internal testing, without releasing the underlying percentage. No independently verified efficiency or accuracy figure has been published; the claim relies on CNIPA's own statements at official briefings.
Implementation
Indicative cost
High (€500k–€5M) — No budget figure is published; classified as high cost given a national in-house AI ecosystem built and progressively expanded across an entire patent office over several years.
Time to results
Medium (1–3 years) — Built from 2022, launched nationally in January 2023, and expanded with further AI tools through July 2025 (about three years).
Staffing & skills
CNIPA patent examiners (retain final decision authority under the Patent Law)
Conditions for success
Explicit human-oversight rule: AI output cannot be used directly as an examination opinion
Iterative expansion of the tool suite (translation, document comparison, large-model search, legal assistant) reaffirmed at successive official press conferences
Common failure modes
No independently verified efficiency or accuracy figure has been published; the reference-detection-rate improvement claim relies solely on CNIPA's own internal testing
Commonly funded by
National / regional programmes
Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.
Do you run this practice?
Claim it —
verified implementers get a public contact pathway and can propose corrections.
Data sources
Where this practice's information was retrieved from, and when.
MLIT runs a national catalog certifying AI and sensor-based bridge/tunnel inspection technologies against published performance criteria. Listings grew from 12 …
CARB flew AI-assisted imaging spectrometers over 22,000 sq. miles of California, examining 272,000+ facilities and finding under 0.2% of infrastructure …