Japan urges AI firms to disclose training data to combat IP infringement

The Japanese government has unveiled draft regulations for generative artificial intelligence (AI) operators, with a focus on intellectual property rights protection. Although not legally binding, these guidelines urge operators to be transparent about their data collection methods and, under certain circumstances, respond to disclosure requests from rights holders and users. The draft rules, derived from a law passed last year, received broad approval during an expert panel meeting in Tokyo.

Companies are expected to publish on their websites a summary of the programs they utilize to automatically gather data from the internet. They are also encouraged to refrain from collecting data from pirate sites. Furthermore, operators will be requested to disclose to rights holders in manga, music, and film, as well as to users generating content with AI, whether copyrighted works are included in their training data.

This disclosure would be necessary, for instance, when a rights holder identifies AI-generated content online that resembles their own work and initiates legal proceedings.

Additionally, operators will be asked to establish systems that inform users if copyrighted works similar to their generated content were part of the training data. The guidelines explicitly state that mandatory disclosure will not be sought for information related to trade secrets or security. While noncompliance carries no penalties, companies will be required to explain their reasons for not adhering to the guidelines.