Anno Calls for Domestic Cybersecurity AI

Tak Hirohiro Anno, leader of Team Mirai, said during a livestream on the party’s official YouTube channel on November 3 that foreign-developed open-weight models could be deliberately weakened or leak data even when operated domestically. He called for domestic model development. Hosting a model on domestic servers and securing data sovereignty still leaves a risk that the model will behave in logically undesirable ways, he said.

Anno said Japan needs to host foreign open-weight models with strong cybersecurity capabilities on domestic servers and use them for defense. But running foreign models on its own servers does not make them “100% safe,” he said, calling for domestic model development to proceed alongside their use.

In a proposal submitted to the Digital Agency on October 23, Team Mirai asked the government to provide an open-weight model testing environment and computing resources, and to support development of a domestic cybersecurity AI model to strengthen national cyber defenses.

During the livestream, Anno said all three measures were needed to address a series of information leaks and cyberattacks: securing access to foreign closed models, using foreign open-weight models with safety measures, and developing domestic models. The article ends by directing readers to the livestream recording. It does not name specific leak cases or models, or provide performance figures. It distinguishes the proposal’s requests for testing resources and domestic model support from the livestream’s call to pursue three measures in parallel; both are presented as Anno’s and Team Mirai’s positions.

Tak Hirohiro Anno, leader of Team Mirai, said during a livestream on the party’s official YouTube channel on November 3 that foreign-developed open-weight models could be deliberately weakened or leak data even when operated domestically. He called for domestic model development. Hosting a model on domestic servers and securing data sovereignty still leaves a risk that the model will behave in logically undesirable ways, he said.

Anno said Japan needs to host foreign open-weight models with strong cybersecurity capabilities on domestic servers and use them for defense. But running foreign models on its own servers does not make them “100% safe,” he said, calling for domestic model development to proceed alongside their use.

In a proposal submitted to the Digital Agency on October 23, Team Mirai asked the government to provide an open-weight model testing environment and computing resources, and to support development of a domestic cybersecurity AI model to strengthen national cyber defenses.

During the livestream, Anno said all three measures were needed to address a series of information leaks and cyberattacks: securing access to foreign closed models, using foreign open-weight models with safety measures, and developing domestic models. The article ends by directing readers to the livestream recording. It does not name specific leak cases or models, or provide performance figures. It distinguishes the proposal’s requests for testing resources and domestic model support from the livestream’s call to pursue three measures in parallel; both are presented as Anno’s and Team Mirai’s positions.