Japan to require AI firms to disclose training data

A government panel in Japan has largely endorsed a proposal to implement a so-called "principle code" for generative artificial intelligence firms. The aim is to safeguard intellectual property rights by mandating companies to openly disclose their AI training data and methods used to collect that data to the public. This draft code emerged from an online session of an expert panel discussing intellectual property rights in the age of AI.

It was formulated in light of mounting apprehensions that texts and images might be utilized for AI model training without proper consent, potentially violating copyrights, as generative AI gains widespread acceptance. The plan is rooted in legislation on AI-related technologies adopted in May 2025, aiming to strike a balance between protecting rights and fostering innovation.

The government will adopt a "comply or explain" policy, outlining a nonbinding code for generative AI businesses, encompassing system developers and service providers. Companies opting to comply will publicly announce their adherence on their websites and inform the government. Even foreign entities offering AI services within Japan will fall under this approach.

The draft code consists of three main principles. The first mandates businesses to disclose the generative AI models they employ, their training data, and methods for acquiring such data on their websites, though sensitive information won't be compulsory. The second principle requires businesses to reveal the inclusion of specific webpages in training data upon request from copyright holders and other rights holders alleging infringement.

The third principle stipulates that businesses must address inquiries from system and service users regarding potential copyright infringements.