Skip to content
Japan
File

Japan: deploying government AI Gennai

This dossier follows Japan’s Gennai government AI trial, its applications, evaluations and human oversight requirements.

Last updated : 2 key moments

Where things stand

As of 30 October 2026, the Digital Agency has published a survey of 6,439 staff and opened a parliamentary-answer drafting prototype. Time savings are self-reported; the prototype remains under evaluation.

Japan opens AI prototype for drafting parliamentary answers

Japan’s Digital Agency announced on 30 October 2026 that about 180,000 staff could access a prototype helping draft government answers to Parliament. The tool is integrated into Gennai, the government’s AI platform.

Japan Digital Agency logo
Japan’s Digital Agency logo, used to identify the institution. — Digital Agency, Japan / Wikimedia Commons — public domain (PD-textlogo).

The system helps find and organise documents and previous answers, then generate draft text. It is to be tested during the extraordinary parliamentary session, with full deployment planned for January 2027, according to the agency.

Development involves 102 evaluators from 16 ministries. Staff must compare each proposal with source documents, check facts and consistency with government policy. Decisions and final approval remain with humans; ministers retain responsibility for their answers. The accuracy of generated text is not guaranteed.

Sources 3

Japan: Gennai users report saving about 110 minutes a week

Japan’s Digital Agency published a survey of its Gennai AI platform on 30 October 2026, with responses from 6,439 government staff. It estimates time savings of 1.83 hours, or about 110 minutes, per user per week.

Japan Digital Agency logo
Japan’s Digital Agency logo, used to identify the institution. — Digital Agency, Japan / Wikimedia Commons — public domain (PD-textlogo).

Among users, 54.4% report saving at least an hour on the tasks concerned and 71.7% say they are satisfied. However, 31.2% of respondents rarely or never use Gennai. Staff also report inaccurate answers and difficulties using it.

The average saving is calculated from self-reported time ranges: it is neither independent time tracking nor a demonstrated reduction in every employee’s working hours. The agency plans training, cross-ministry sharing of tools and another survey around January 2027.

Sources 3