01
Application context & facts
- The Shenzhen Intermediate People's Court developed an AI-assisted adjudication system and deployed a vertical model for judicial work.
- The application report republished by the Greater Bay Area portal on 27 April 2026 stated that the system had assisted with more than 600,000 cases at that time. This is a dated reported figure, not an independent audit or live count.
- The project was listed among Guangdong's representative AI-plus application cases.
02
Core legal & governance issues
- How judicial data entering models and knowledge bases should be protected and access-controlled
- How generated suggestions, similar-case search, and drafting outputs should be source-verified and reviewed
- How models, prompts, knowledge bases, and logs can support judicial accountability and audit
03
Reported application
- Public reporting describes a specialized judicial model and case-handling assistance.
- The project was included among Guangdong’s representative AI applications for legal and judicial work.
- The disclosed material describes construction achievements; architecture, measured performance and implemented procedures require project documentation for verification.
04
Practical compliance guidance
01Judicial and legal systems should retain an assistive role and require authorized personnel to review facts, authorities, and final outputs.
02Case data used for training, retrieval, and generation requires classification, least-privilege access, and complete logs.
03Acceptance testing should cover accuracy, citation traceability, unauthorized access, bias, hallucination, and fallback procedures.