AIBRIEF
Non binding "principle code" asks generative AI operators to publish how they collect training data and, on rights holder request, confirm whether specific copyrighted works sit in their training sets.
Japan's Cabinet Office broadly approved a draft "principle code" on Aug. 18, 2026 that asks generative-AI operators to publish how they collect training data and to confirm to rights-holders whether specific copyrighted works are in their training sets. The rules are not legally binding, and the government has signaled finalization "in the near future" without naming a date.
The code is grounded in Japan's AI-related technology law enacted in May 2025, according to Kyodo News via Japan Today and the Japan Times. The Mainichi Shimbun and Nippon.com, drawing on Jiji Press reporting, said the framework was endorsed at the Cabinet Office expert panel on IP rights in the AI era. The panel's meeting record is the primary regulatory source.
Operators are asked to publish overviews of programs they use to automatically collect data online and to avoid scraping from pirate sites. Upon request, they should disclose to manga, music, and film rights-holders, and to users, whether copyrighted works appeared in training data. The disclosure trigger includes a rights-holder finding AI-generated content resembling their work and preparing litigation. Operators are also expected to build systems that tell end users whether similar copyrighted works were in the training set.
The draft's bite is the public expectation that noncompliant companies explain why, on the record, rather than any fine. Trade-secret and security information is carved out: the draft states "compulsory disclosure will not be sought" for that data, to avoid pushing operators to disclose more than the underlying law allows. Overseas operators offering generative-AI services in Japan are in scope, not just domestic players, per Nikkei and Yomiuri.