OpenAI explains data disclosure limits under Japan's generative AI principles code
OpenAI解释日本AI原则下数据披露限制
OpenAI published its disclosure made under the "Code of Practice for Intellectual Property Protection and Transparency in the Appropriate Use of Generative AI, etc." of the Intellectual Property Strategy Promotion Office, Cabinet Office of Japan, dated August 25, 2026, updated on November 13, 2026; the disclosure excludes trade secrets and security-sensitive information. Under Principle 1, OpenAI lists measures including non-confidential architecture and training process information in system cards, the training data summary published under the EU AI Act, explanations of crawler behavior such as GPTBot, and the G7 Hiroshima AI Process transparency report, and states that its crawlers follow robots.txt, do not bypass CAPTCHAs or paywalls, refer to the USTR Notorious Markets List to avoid using content from piracy sites in training, and label content provenance with C2PA and SynthID. For the disclosure frameworks of Principles 2 and 3, OpenAI explains under the code's comply-or-explain mechanism that it cannot implement them, stating that details of training and validation data are trade secrets, that disclosure would help competitors learn about its training datasets, and that courts in other jurisdictions have already recognized their secrecy, and that the models are not designed to store or reproduce specific copyrighted works.