Summary
- Training, retrieval, quotation and generated substitution create different evidence and licensing questions.
- Publishers and AI services need source logs, permission records, removal handling and a workable challenge process.
The CNKI and Metaso AI dispute illustrates how quickly search, summarisation and generation can blur into one product experience. A claimant needs more than similarity, and a provider needs more than a general assurance about lawful data. Systems should preserve source identifiers, retrieval events, licence scope and the output shown to the user, while protecting confidential material. The next useful evidence is a reproducible case file for one challenged answer from ingestion to display and correction. Traceability will not settle every legal question, but without it neither negotiation nor adjudication can begin from shared facts.
