Japan's プリンシプル・コード lets firms query AI training data from Oct 26

The Culture Agency's copyright subcommittee laid the groundwork on March 15, 2024, splitting AI and copyright into three stages: development and training, generation and use, and whether AI output itself counts as a work. It followed on July 31, 2024 with a checklist and guidance aimed at business users, which industry can translate directly into internal procedure. One point deserves attention: if an existing work was part of an AI's training data, the agency's summary says reliance on that work is normally presumed even when the user did not know of it. Generating material internally may fall within private-use limits, but distributing it online or in print often sits outside those limits.

The Principles Code adds a different mechanism, aimed at the provider side rather than the user. Its third principle lets a person who created something with generative AI ask the provider whether content resembling their output is in the training data. The catch is that meeting those conditions requires a saved prompt, which ties the government's transparency mechanism to everyday record-keeping habits inside companies.

Separately, the agency's view on whether AI output is itself protected cuts both ways. Output generated autonomously, or from a simple instruction, is not considered a work, while use of AI as a tool for expressing thought can qualify. In a related matter reported by Kyodo News, Chiba prefectural police referred a man from Kanagawa to prosecutors on suspicion of copyright infringement on November 20, 2025, over AI-generated images copied without permission for a book cover. Police judged the images to be works because they were made by repeatedly giving the AI specific instructions, a first for such a referral, though this is a police view and not a court ruling.