Japan Targets AI’s Black Box
Tokyo wants AI companies to reveal more about the data behind their models, as the global battle over copyright and AI training data intensifies
Japanese Prime Minister Sanae Takaichi arrives at her office in Tokyo, Japan, on August 27, 2026. Photo credit: Kaname Yoneyama / The Yomiuri Shimbun via Reuters Connect
Japan is moving to increase transparency around the data used to train generative AI models.
This week, the government adopted new principles calling on generative AI companies to disclose the types of data used to train their systems, how the data was collected, and the training process. The framework is not legally binding, but companies that adopt it will be required to publish details about how they operate.
The principles also apply to foreign AI companies whose systems or services are available in Japan, meaning the initiative could have significance beyond the country’s borders.
The move comes as copyright has become one of the key unresolved issues facing the global AI industry. Model developers need vast amounts of text, images and other content to train increasingly sophisticated systems, while publishers, artists and other rights holders are demanding greater transparency about the use of their work.
Under the framework, companies will be expected to respect restrictions such as paywalls, comply with website instructions such as robots.txt, and refrain from scraping piracy sites. They will also be asked to maintain records related to training processes and provide channels through which rights holders can request information.
The framework does not require companies to disclose their complete training datasets. Instead, it seeks to increase transparency while protecting trade secrets.
Japan’s approach is particularly interesting because the country’s existing copyright law already permits certain uses of copyrighted works for data analysis, including AI training, under certain conditions. The new principles therefore focus on how AI companies collect and report data, rather than on fundamentally changing copyright law.
The initiative comes at a time when governments around the world are grappling with similar questions about training data and copyright.
At the same time, Japan is seeking to accelerate AI adoption. A Reuters survey published earlier this month found that more than 80% of Japanese companies are using AI only to a limited extent or have yet to adopt the technology. The challenge for Tokyo is to encourage an industry it sees as strategically important while giving creators greater transparency into the data powering it.
The big question now facing the global AI ecosystem is no longer simply what AI can produce — but who owns the data that enables it to do so, who controls it, and who gets paid for it.