Skip to main content

This page has been translated using TexTra by NICT. Please note that the translation may not be completely accurate.
If you find any mistranslations, we appreciate your feedback on the "Request form for improving the automatic translation ".

Text: Guidebook on Measures against Risks in generative AI Utilization (α version)

Based on the Priority plan (decided by the Cabinet on June 7, 2022) for the realization of the digital society, Digital Agency is investigating the appropriate use of AI in the public and private sectors after understanding the actual situation and trends of AI and identifying risks and necessary measures.

Recent technological innovations in generative AI and other areas have made it possible to obtain various advantages. While assessing the trend of such technologies, the government has repeatedly discussed the business use of generative AI at related ministries and agencies at the 10th Secretariat Meeting and the Written Document Holding Council for the Promotion of a Digital Society. In addition, since December 2023, we have conducted a technical validation for the appropriate use of generative AI and released the results *.
*For details on the results of the technology validation, please see the in which the technology validation for the appropriate utilization of generative AI in Digital Agency and administration in 2023 was implemented.

Based on the history of the discussions and the results of the validation, we will publish the "Guidebook on Measures against Risks in Use of Text generative AI (alpha version)." This is a technical report on the risks that can be assumed when using generative AI in public service, etc. and the measures to be taken.
At present, there are some insufficient points in the content because it is an alpha version, but please use it as a reference for discussions on risks and measures to be taken when using generative AI. If there are any missing descriptions or inaccurate expressions in the alpha version, please contact with your opinions and requests. In the future, we will regularly expand and improve the content for publication as an official version. Please contact .

Disclosure Background

When using generative AI, there are guidelines for administrative work such as the Hiroshima AI Process (Ministry of Internal Affairs and Communications) and AI Business Operator Guidelines . Based on these guidelines, the "Guidebook on Measures against Risks in Text generative AI Utilization (α version)" focuses on Text generative AI and its utilization in generative AI, and describes more specific risks and measures to be considered by the Government procurement.

Features of the Guidebook on Measures against Risks in Text generative AI Utilization (α version)

Targeted at text generative AI service development operators and providers

Generative AI can generate not only text but also images, sound, and video. Since much of the information exchanged in actual business is text information, and the utilization of generative AI other than text information in Digital Agency is at the validation stage, we focus only on text generative AI in generative AI.
In addition, it is assumed that the information will be used mainly by persons engaged in work related to system generative AI for utilizing development. Therefore, the information does not include the risks of those who development the infrastructure model (such as risks related to the supply chain of learning data and rights issues) or the risks of service users (such as risks related to deepfakes).

Classification of text generative AI by mode of use, assumed use cases, and process

The assumed risks of text generative AI differ depending on the form of use (such as the state of being incorporated into a specific service or the Web API format for using a large-scale language model) and use cases. In the alpha version, we are aware of the differences in the form of use and use cases, and list the risks and points to be noted according to the process defined in the Digital Government Promotion Standard Guidelines Practice Guidebook.

Describe the risks and precautions identified by the validation as much as possible.

Assumed risks discovered in the course of validation are described as much as possible. In the future, we will continue to update it and aim to publish it as an official version of the "Guidebook on Measures against Risks in Use of Text generative AI."

Target audience for the "Guidebook on Measures against Risks in Utilization of Text generative AI (α version)"

It is intended for administrative staff involved in the introduction of generative AI into government information systems. Please use it in conjunction with 's "Practical Guidebook for Digital Government Promotion Standard Guidelines" and 's "AI Business Operator Guidelines (Version 1.0)" .
In addition, even if you are not an administrative official, if you are considering introducing Text business improvement for your own services or generative AI, please use it as a reference to consider possible risks and measures when using generative AI.

Composition of the "Guidebook on Measures against Risks in Utilization of Text generative AI (α version)"

It consists of a main text (the first six chapters) and an appendix (the second three chapters) with advanced content.

Positioning and Roadmap

We plan to consider incorporating it into the digital society Promotion Standard Guidelines by the end of fiscal 2024.

Download Publications

Text: Guidebook on Measures to Reduce Risks When Using generative AI (α version) (PDF / 1,742 kb) (updated on June 10, 2024)

Updates

  • First published on May 29, 2024
  • May 30, 2024 Minor corrections to clerical errors, etc.
  • June 10, 2024 Minor corrections to clerical errors, etc.