The Spread of Generative AI Use in Japanese City Halls: The True Meaning of "No Information Will Be Disclosed Externally" and the Reality of Staff's "Temperature Gap"

In Japanese city halls, the use of generative AI for official tasks is rapidly expanding. What began as grassroots experimentation has now evolved into secure implementations that integrate APIs with internal chats, the creation of national guidelines, and the sharing of use cases. Notably, there are many cases where it is explained that "no information will be leaked externally," but this is not an "unconditional safety declaration." This often refers to a set of operational rules: ① use within closed networks such as LGWAN, ② settings that prevent secondary use (learning) of input data, and ③ prohibitions on inputting confidential or personal information. On the other hand, there remains a "temperature difference" at the operational level due to differences in departmental business characteristics, information management levels, and digital skills. Proper education, governance, and visibility of outcomes are key. In advanced municipalities, tasks such as creating greeting messages, notifications, meeting notes, FAQ drafts, translations, and reading scripts are being optimized for both time efficiency and quality improvement. This trend also supports responding to inquiries from foreign residents and multilingualization. This article explains the actual mechanisms, risks and limitations, success stories, and how to overcome the "temperature difference" in a way that overseas readers can understand. [Sources are referenced in the main text and at the end.]