Japan asks AI firms to disclose training data

Japan adopted a voluntary Principles Code asking domestic and overseas AI providers serving the country to disclose their training processes, data types, and collection methods, according to Jiji Press’s report on the government’s code. Companies that accept all or part of it must notify the government and publish the information; under conditions, they must also say whether training data could contain material that infringes copyright. The three requested disclosures give model vendors a compact provenance checklist. The code is not legally binding, but model vendors should treat the three requested disclosures as a preview of the provenance records buyers and rights holders will increasingly demand, extending the archive’s analysis of runtime content provenance.