On August 25, Japan published a Principle Code for intellectual-property protection and transparency in generative AI, setting out principles for model and service providers around training data and inquiries from rights holders.
This is not a binding law
The document is not a law and does not create direct penalties. It uses a comply-or-explain approach: providers can follow the principles, or explain publicly why they do not.
It covers companies that develop generative AI models and providers that integrate them into products or services. Overseas companies can also be covered when they offer AI services in Japan.
Three core principles
First, providers should disclose model architecture, licensing, training methods, types of training data and measures taken to protect intellectual property. This does not require publishing every training item; it asks for a basic account of the model’s sources and handling.
Second, providers should respond when rights holders identify a URL and ask, under conditions such as legal proceedings, whether their work was used for training. Third, users can also inquire when they find a similar work connected to content generated by an AI service, subject to stated conditions.
What changes for creators and AI companies?
For creators, this does not guarantee that every inquiry will receive complete data. It gradually turns the question of how to ask and how providers respond into a process AI services are expected to prepare.
AI companies will need stronger records of model versions, data types, licensing status and inquiry handling. If they do not disclose some information, they will need a public explanation.
What happens next?
Japan says providers that accept the Principle Code will notify the government and publish related information, while the start date for notifications will be announced separately. It should therefore be understood as a voluntary governance framework, not an already effective mandatory data-disclosure law.
