September 19, 2025
TOKYO – The government plans to support the development of a domestic artificial intelligence model, based on Japanese data and technology, in a bid to avoid reliance on AI from the United States and China, according to sources.
The United States and China have surged ahead in the development of generative AI, which can create text and other content, but there are concerns that relying on these overseas models could result in data flowing abroad and incorrect information about Japan being widely disseminated. This has been considered a problem from a security perspective.
Resources will be provided to Japanese companies for the development of learning data and other features, to support the creation of what the government hopes will become a highly reliable domestic AI model.
The National Institute of Information and Communications Technology (NICT), which is under the jurisdiction of the Internal Affairs and Communications Ministry, will provide data in Japanese collected over the past about 20 years. NICT and Preferred Networks, Inc., a Tokyo-based AI development company, will jointly develop an AI model that generates highly reliable answers on Japan’s culture, customs, systems and other fields.
The government anticipates that information technology company Sakura internet Inc. will provide the new domestic AI model through data centers located in Japan.
The communications ministry and the Economy, Trade and Industry Ministry plan to provide financial and other support for such matters as the provision of infrastructure necessary for the model’s development, the recruitment of talented engineers possessing exceptional abilities and the maintenance of high-quality Japanese-language data. It is expected that the central government, local governments and companies will use the Japanese AI model.
Japanese firms and organizations will handle every aspect from the provision of data used for AI learning, the development of a “large language model” that will become the model’s brain, and operation of the data centers, as the government aims to create and provide a completely domestic generative AI model.
Use of U.S. and Chinese AI models has been growing in Japan. However, the learning data and other elements of these overseas models’ development processes are unclear, which has sparked concerns among some Japanese government officials. “There are fears that these models learn data that’s unacceptable from Japan’s point of view,” a government source said.
Overseas models learn data mostly from English-language sources, which some observers have said have insufficient understanding of Japan’s culture, history and other fields. Another problem is that the principles and opinions of the nation that developed those AI models can influence the answers they generate.
In one case, Liberal Democratic Party Policy Research Council Chairperson Itsunori Onodera asked China’s DeepSeek generative AI model if the Senkaku Islands in Okinawa Prefecture were Japanese territory. DeepSeek answered that the islands were “an inherent territory of China.”
Believing that dependence on overseas AI models presented some risk, the government decided to support the development of an AI model that would deliver highly trustworthy answers base on Japan’s culture, customs, history and other fields.
AI development requires vast amounts of data, computers and excellent human resources with specialized knowledge. However, Japan is lagging the field leaders United States and China, which have immense development resources, in this regard.
According to the Artificial Analysis Intelligence Index, an international composite metric for assessing generative AI capabilities, AI services from the United States and China, such as OpenAI and Alibaba, dominate the field. A growing number of Japanese companies also use these U.S. and Chinese AI models as a base, which has raised concerns about relying on such services from overseas.