The Japanese government recently adopted a set of guidelines for generative AI providers known as the “Principles Code.” The move comes in response to concerns regarding infringement of intellectual property rights (source: The Nikkei).
The guidelines center on urging gen-AI providers to disclose the data used to train their AI models (and how they acquired it) to ensure greater transparency. The guidelines are not legally binding, so companies are essentially expected to take voluntary action.
The government plans to begin putting the Principles Code into practice and accepting responses from providers around this fall. Companies that refuse to comply will be asked to explain why. Notably, the guidelines will target not only companies headquartered in Japan, but also those offering AI systems or services targeted at the Japanese market.
As for businesses that do comply with the new guidelines, the government will publish a list of their names, along with links to the relevant information disclosed on their respective websites. This includes information on their generative AI training processes, the types of data used for training, and how that data is collected.
The Principles Code also stipulates that if requested by copyright holders or users, AI service providers will disclose (given certain conditions are met) whether copyrighted works or other protected materials are included in their training data. This is intended to create a system that allows people to determine whether copyright infringement may have occurred. At the same time, the guidelines also state that they “do not require the mandatory disclosure” of “sensitive information” concerning trade secrets or safety.
Reponses to the new measures among Japanese users have been varied. While many are happy to see efforts being made to prevent IP infringement, there is also doubt about whether it is realistic to expect businesses (especially major overseas AI companies) to comply with the guidelines given the lack of legal enforcement or punitive measures.
