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 ".

The 3rd Advanced AI Utilization Advisory Board

Last Updated:

Outline

For an overview of the Committee, please see Advanced AI Utilization Advisory Board .

Event Information

Date:
Tuesday, March 10, 2026, from 9:30 a.m. to 11:30 a.m.
Location:
Office of the Prime Minister's Office - Online

Proceedings

  • 1. Opening
  • 2. Proceedings
    • 1. Outline of Periodic Report on generative AI System by Each Ministry and Agency
    • 2. Trends in generative AI in Japan and other countries
    • 3. Proposed Revisions to Enhance the Guidelines Concerning the Acquisition and Utilization of generative AI for the Evolution and Reform of Public Administration
  • 3) Closing

Data

Reference Documents

Attendee

(1) Members
Chairman Kadobayashi, members Okada, Kitamura, Chai Shan, Torisawa, Naganuma, Nabatame and Yoshinaga
(2) Digital Agency
Matumoto Minister for Digital Transformation, Kawashima Parliamentary Vice-Minister for Digital Transformation, Asanuma Consultant, Itoh Consultant, Misumi Vice-Minister for Digital Transformation, Chief Officer of Digital Agency, Mr. Tomiyasu, Vice-Minister for Digital Policy, Ogiwara Director-General, Hasui Director-General, Morita Administrative Deputy Director-General, Ibata Deputy Director-General, Okuda Deputy Director-General, Kitama Deputy Director-General, Naito Director, Hasimoto Director, and Yamaguchi Director

Summary of proceedings

At the beginning, Mr. Parliamentary Vice-Minister for Digital Transformation Kawasaki stated that as the use of AI agents increases with the emergence of robotics and stablecoins, it is important for this plenary session to enhance the guidelines based on the "Basic Plan for AI."

(1) Outline of Periodic Reports on generative AI System by Each Ministry and Agency

The Secretariat made a report in Appendix 1 on the results of an analysis of the state of generative AI utilization by the various ministries and agencies.

The main questions and opinions of the attendees are as follows.

Nabatame Member: I understand that each ministry and agency is making progress in the practical application of periodic reporting, and that qualitative and quantitative results are visible. I believe that we are beginning to see some areas that can be organized as quantitative performance targets. From a medium - to long-term perspective, I believe it will be effective to make efforts to share with the public the results each ministry and agency has achieved through the introduction of AI. I hope these efforts will contribute to the efficient operation of each ministry and agency. Private companies have also confirmed that the introduction of AI has reduced working hours, but in some cases, the time saved in the organization has been absorbed. This does not mean that the number of workers has been reduced as a result, but there are cases where people simply work overtime until 8 p.m. and finish it by 5 p.m., but the actual state of work does not change. I believe that issues that are made visible in this way are important for considering the operational efficiency of each ministry and agency.
Secretariat: : Regarding the outcome targets, we have just started to promote AI utilization this fiscal year. First of all, we have set up indicators while observing how much business efficiency can be improved. We will continue to pay attention to actual utilization and how the outcome targets can be achieved. As in the private sector, we have observed cases where the utilization of AI has not reduced business hours. Since there may be situations that cannot be solved by government offices alone, we would like to refer to good management schemes in the private sector if there are any.
Secretariat: AI and consider how to present them. In terms of promoting utilization, it is important to widely share the benefits that can be seen from objective indicators. With that in mind, we will continue to consider how to set indicators and how to present performance targets.

Member: Yoshinaga: AI must bring benefits to government officials, such as operational efficiencies, so it is significant in terms of making the results visible, and I think it is a good initiative. In addition, the risk differs depending on the purpose of use. For example, a system for the public is particularly risky. If generative AI is to be utilized for policy planning and formation within each ministry and agency in the future, it is necessary to be careful not to make judgments based on wrong information. It is necessary to collect information on the purpose of use in the future. I would like to ask you to show the details of the purpose of use, including whether it is just for collecting information or whether it is used for policy planning.
Secretariat: use. We would like to continue to organize matters such as what kind of classification can be made, and consider sharing them with various ministries and agencies.

Chai Shan Members: Material 1, it is stated that the use of AI that refers to Confidentiality class-2 information is increasing, and we believe that this is solely the result of Digital Agency's efforts. Regarding such use, for example, in the case where a government official wants to use AI for meeting minutes, is there a mechanism or system for responding to inquiries and inquiries addressed to Digital Agency, such as whether there is a precedent, what kind of service is used under what conditions, etc.? Recently, when I was personally consulted by a national corporation, they told me what kind of precedents there are specifically, and that it would be reassuring if there were precedents. Is there a mechanism for sharing the presence or absence of specific precedents and their contents within the government?
Secretariat: Digital Agency As the secretariat, we have opened a AI consultation window, and we are operating a mechanism to link it to the departments and agencies in charge. Although the accumulation of use cases is still in progress, it is possible to share examples of utilization according to the content of consultation at this stage.
Chai Shan Members: To a certain extent, does the organization respond to inquiries regarding specific information, such as the services being used?
Secretariat: You are right. Depending on the case, we may hold hearings on the specific contents of the case, and we also collect specific information through hearings.

TORISAWA: The information in the summary of the periodic report is coarse grained, and it is hard to imagine what measures should be taken in the future. It may be difficult to collect more detailed information because it will be a burden on each ministry and agency. However, it would be good to collect information on the purpose of use and what kind of system it is, even if only briefly. I understand that each ministry and agency is promoting the use of the system. I think it is good to move from the stage of totaling the number of units to the stage of analyzing them.
Secretariat: Since we believe that the necessary information changes according to the situation, we would like to review the content of the analysis while taking into account the needs of each ministry and agency. In particular, we would like to continue to review the information on the purpose of use and what systems and models are used.

(2) Trends in generative AI in Japan and other countries

The Secretariat reported the trend of generative AI in Japan and other countries in Appendix 2.

The main questions and opinions of the attendees are as follows.

Nabatame Member: AISI's "AI Incident Response Approach Book" states that visualization (observability) and containment (controllability) are important, which is true. There is a question as to who will implement visualization and how. The Singaporean "Model AI Governance Framework for Agent-Based AI" states that humans should take responsibility, and we believe that how humans implement the process of visualization is important. I would like to hear what kind of framework Digital Agency is considering and what kinds of efforts are being made by each ministry and agency.
Secretariat: : With regard to visualization (observability), the "AI Incident Response Approach Book" is largely intended for visualization in terms of security, and Material 3 also describes measures to make it possible to verify by taking logs, etc., in the Guidelines. As for accountability, AI Guideline for Business mentions human-centered, and the Guidelines for Utilization of AI Procurements also mention accountability. It is clear that the administration must be primarily accountable. Regarding the visualization of the "AI Incident Response Approach Book", is there any additional information from member Kitamura?
A member of KITAMURA: Secretariat, the framework of the "AI Incident Response Approach Book" first focuses on cybersecurity. We call it an isolated zone, but it is important to securely isolate AI and continue to use it safely and securely in the event of an incident while using AI agents, etc. AISI plans to release a practical guidebook on CAIO within this month. We believe it is important to consider how to update once established rules and guidelines based on future trends.

Member: Yoshinaga: Agentic AI model In addition to the AI governance framework, the OECD has also released a report on definitions for AI Agent/Agentic AI, which we hope you will refer to. The US NIST has also announced efforts to establish technical standards for AI agents, which we hope you will refer to.
Secretariat: : The OECD's definition of "agentic AI" is also mentioned in AI Guideline for Business's revised proposal, but it is still under consideration and is therefore provisional. We are considering referring to that definition in the explanatory notes, etc., when AI Guideline for Business's definition is finalized.

Member: : Looking at Exhibit 1, I think that in Japan, more people use AI for efficiency and cost reduction. On the other hand, in Exhibit 2, foreign countries' trends emphasize efficiency and safe use. In the German "Guiding Principles Based on Values as a Common Basis for Action," the positive term "chance-oriented" is used. I would like you to investigate whether this "chance-oriented" is assumed to be used to increase added value or to discover new outcomes, or whether it is just for efficiency improvement. With reference to that information, I would like you to consider how to use it positively in Japan.
Secretariat: : There are concerns that the use of AI itself is becoming a purpose. We believe that it is necessary to consider not only the utilization rate but also the purpose of using AI and what can be realized by using it. As a tendency seen in the cases of IT failures, there were cases where the use of IT itself became a purpose. Since we believe that "chance-oriented" is also related to such a perspective, we would like to continue to consider it while conducting thorough research.

Chairman: Kadobayashi: Through the discussions on Agenda 1 and 2, there is room for further discussion on the purpose of AI utilization and observability. Therefore, we would like to continue to look at best practices in industry, government, and academia, and consider methods for grasping the current situation.

(3) Proposed Revisions to Enhance the Guidelines Concerning the Acquisition and Utilization of generative AI for the Evolution and Reform of Public Administration

The Secretariat reported on the proposed revisions to improve the guidelines in Appendix 3.

The main questions and opinions of the attendees are as follows.

Naganuma: , Member: The first point is, regarding the target AI of the Guidelines, I support the direction of expanding the scope to include the Keidanren related companies with audio and image outputs. I believe that it is suitable for the actual situation of utilization of both the public and private sectors. The first point is, regarding the target data of the Guidelines, I support the direction of expanding the scope to include the generative AI with audio and image outputs. I believe that it is suitable for the actual situation of utilization of both the public and private sectors. The second point is, regarding the risk judgment logic, there was an update this time, and I evaluate it in that it is based on the concept of security design and is consistent with the practical sense of private companies. On the other hand, I am concerned that in the operation stage, if the judgment of what is considered to be a significant impact is too much left to individual CAIO, there will be variations among ministries and agencies and projects. The third point is, regarding the procurement check sheet, I believe that it will lead to the improvement of transparency and accountability by enriching the examples of measures, supporting information, benchmarks, etc. It is stated that "as much as possible" for the examples of measures, but I believe that it will lead to discussions on the combination of certain requirements with existing mechanisms such as external certification to make it realistic from the viewpoint of competition. The fourth point is, regarding the current operation, I believe that it will need to refer to the improvement of the risk judgment logic, I agree that it is based on the scope. I believe that it is based on the concept. I believe that it is consistent with the practical sense of the private sector. I am concerned. On the second point. I is based on the approach, I am concerned with the practical sense of the private sector. I am concerned with the scope, I am concerned. I am concerned. I I. I. I. I. I am concerned. I. I. I. I. I. I. I. I I. I. I. I. I. I. I. I. I. I. I. I I. I. I. I. I. I. I. I. I. I. I. I. I. I am concerned I. I. I. I. I. I. I. I. I. I. I. I. I. I. I I. I. I. I am concerned. I. I. I. I. I. I. I. I. I. I. I. I. I. I. I believe. I. I. I. I. I. I. I. I. I. I. I. I. I. I. I. I. I. I. I. I. I. I. I. I. I. I. I. I. I believe I. I. I. I I I. I. I. I I. I. I. I. I. I. I. I. I. I. I am concerned. I'm concerned. I am concerned. I. I am. I. I. I I.
Secretariat: Regarding the second risk logic, there are still few high-risk projects. We assume that the number will increase as the operation progresses, but we consider that the elaboration based on specific examples is an issue to be considered in the future. We receive consultations from each ministry and agency at the AI consultation window. We consider that the elaboration based on specific examples is an issue to be considered in the future. As for the contents related to the third point, regarding the learning method, there is a description of observability on page 24 of Appendix 3, but at least it describes the contents that should be known when using AI, such as "acquiring information such as main performance indicator information." In addition, it states that "to the extent reasonable for risk analysis and consideration of risk response." We will strive to ensure that the part related to trade secrets in generative AI development will not be required within the necessary scope. We will consider the viewpoint that the requirements of the procurement check sheet are excessive, taking into account the results of soliciting opinions in the future. Regarding the utilization of external certification, we have decided not to utilize it at the present time, but we will consider it while observing the actual introduction situation. Regarding the fourth point, we have incorporated necessary contents in the guidelines from the viewpoint of AI and utilization, while cross-referencing related intra-government and procurement SI guidelines. Related guidelines are described in the procurement check sheet, so that at least a comprehensive grasp is possible in the procurement and utilization.

Member: Yoshinaga: Regarding the amendment of "6.5 Governmental Actions to be Taken by generative AI System Providers" on page 14: "In relation to the generative AI System that enables users to create AI agents to perform advanced tasks, in the case where the Electronic Data Processing System that enables users to perform tasks without judging the appropriateness of the output result has been created, it is required to be reported to the provider or the AI Administration Officer (CAIO)." We believe that the Terms of Use should also state that the Electronic Data Processing System will not be used beyond the scope of use provided by the AI System Provider. What specific events are assumed?
Secretariat: AI It is assumed that it can be linked with various applications such as Agent, and some actions can be automatically performed. When an application linkage is set, replies to emails, etc. are automatically performed, and in some cases, replies can be made without human eyes. Major services are now able to use such services. How much this can be used within the government is a realistic challenge, and the risk is high for parts that are out of the human in the loop. For this reason, when making such a system available for provision, it is required to make a declaration and determine the risk in advance.
Member: Yoshinaga: We are concerned about whether it is necessary to report to the system provider as well. We believe that reporting everything to the CAIO will also be a burden on the CAIO, but we would like to confirm whether such a description is made from the perspective of risk management.
Secretariat: : It is voluntary to report to the donor. However, even if the donor is reported, it should be reported to the CAIO.
Chairman: Kadobayashi: The secretariat will examine the relevant sections, including improvement of writing style.

TORISAWA: You said that you would improve the checklist through public comments. However, since not many operators are developing domestic LLMs, I think it would be better to have direct discussions with operators. There is a concern that public comments alone cannot pick up all the nuances. In addition, it would be better to have discussions on whether or not the requirements of the check items in the checklist can actually be met. I have heard that seven domestic LLMs have been selected in "GENAI." I think it would be good to consider consultations with operators in that field.
Secretariat: Secretariat intends to exchange opinions with businesses in parallel with public comments.
Chairman: Kadobayashi: I am also concerned about this. I understand that it is necessary to closely coordinate with creators such as LLM and AI agents. Does it mean that the secretariat will separately coordinate such an occasion?
Secretariat: I intend to work in this way.

Nabatame Member: I agree with the member's emphasis on the revision of intellectual property rights, and I agree with the fact that a clear description of this matter has been expanded. In the United States, lawsuits related to AI are commonplace, and the most common case is copyright infringement. Based on this, I understand that the description of protection against copyright infringement is made with the starting point of not infringing the intellectual property rights of others. I think there are cases in which each ministry and agency has intellectual property rights, but there are also cases in which a third party accesses government data with malicious intent and infringes on copyrights. I would like to know what measures have been taken and what precautions have been taken against such cases.
Secretariat: We assume that most of the information published by the Japanese government on its website, etc. is provided as open data, etc. For such information, we have established rules for open data since before generative AI. We also assume that the case where AI systems access open data is basically within the scope of rules that have already been developed. Basically, when using data released by the Cabinet Office and ministries and agencies, there is no problem in using it if you describe the source, etc.
Nabatame Member: We understand that it can be covered by the existing rules. On top of that, in generative AI, there is a possibility of malicious use such as fake generation and modification. Therefore, we believe that it would be good to consider issuing a certain alert regarding the use of government data against generative AI.
Secretariat: administration.

A member of KITAMURA: AI are changing rapidly, and there is a limit to the conventional method of revising regulations and guidelines every six months to one year. Therefore, it is important to adopt a method of updating the content more flexibly, such as the form of a so-called release note, as a mechanism for announcing changes and updated content as needed. In addition, it is important to consider the ideal way of operating it in a way that leverages Japan's strengths.
Secretariat: In the guidelines this time as well, consideration was given to a form that would make it easier to update the branches and leaves without changing the trunk. For example, the detailed examples of measures are still positioned as reference information at present, but it is possible that they will be separated from the guidelines in the future. In the points of attention ① (4) of the revised draft, a description regarding the call for attention to CAIO has been added, and in practice, CAIO liaison meetings are also held. In addition to the revision of the main body of the Guidelines, we would like to share information on operational aspects that cannot be covered by the Guidelines alone, using opportunities such as these CAIO liaison meetings.
A member of KITAMURA: I think you are right. The year 2025 is called the year of AI agents, and the year 2026 is called the year of physical AI. In the United States, it is expected that a systematic approach will be introduced in the first and second quarters regarding AI agents in view of demonstration projects. Therefore, it is necessary to keep a close eye on these developments. In addition, I would like to share the situation when I visited a German research institution and university last year, as reference information for a continuous trend survey to understand the meaning of "chance-oriented" in the German "Guidelines on the Use of AI in the Federal Administration" in Appendix 2. In the exchange of opinions at the visit, it was recognized that German industrial competitiveness will be strengthened through the use of digital transformation and AI, which will lead to the enhancement of national power. At that time, there was also a mention of the importance of initiatives that take advantage of Japan's strengths, such as quality assurance. Since quality assurance is also a strength for Japan, the trend of Germany will be a reference for future consideration.
Secretariat: We would like to learn from the German example.

Member: These Guidelines are based on the idea of how to promote the use of generative AI, assuming that generative AI is not used in the first place. Therefore, it is assumed that this Guidelines will be read by humans and used after understanding by humans. In the future, such a premise will change, and I think there will be a trend where generative AI will read these Guidelines and utilize them. There may be cases where the ministries and agencies subject to the Guidelines utilize the Guidelines by having Tokyo read the contents and ask questions, but there is also the possibility that new risks will arise due to changes in the way the Guidelines are used. What kind of examination and discussion are you advancing regarding this matter?
Secretariat: Based on the current procurement check sheet, "GENAI" can also be used to check whether or not the procurement specifications are satisfied. Even in the current guidelines, "generative AI" can be used more appropriately. On the other hand, it is a matter of course that there remains a concern that the risk of hallucination will not be zero when using generative AI. Therefore, it is essential to check with human eyes. In the future, we will continue to consider machine readability and provide it in a friendly form for such utilization methods.
Member: At present, it is clear that human beings are responsible for the output results. However, as the responsibility may become ambiguous based on future developments, we request that you continue to study this matter.
A member of KITAMURA: is rapid. generative AI is moving from the "introduction stage" to the "stage of being widely used in the whole society," and the standardization of technology is also progressing. Considering such changes in the phase of AI utilization, it is not enough to just respond based on the conventional introduction stage. It is important to consider an approach suitable for the new phase based on widespread use.

Naganuma: The private sector also faces similar issues, but there are concerns that CAIO's tasks will become enormous. Global think tanks and others are discussing the governance of AI agents. It is recognized that the method of risk control will change from the conventional AI governance. As what the CAIO must see changes, will the Cabinet Office and each Ministry consider initiatives to change the mindset of the CAIO?
Secretariat: The CAIO Liaison Committee shares the contents of the revised guidelines and knowledge. The CAIO also participated in today's observation. The CAIO and the departments that support the CAIO participate in the Advisory Board to catch up with each other. Regarding the governance of AI agents, I would like to refer to foreign countries and the private sector for the form of discipline that should be as well as making it an object of report even within the administration. Based on the best practices of the ministries and agencies that are promoting the utilization of AI agents, I would like to consider the deployment to the ministries and agencies as appropriate.
A member of KITAMURA: AISI is accelerating the research project on AI agents and will publish the evaluation viewpoint guide on AI agents next year.

Parliamentary Vice-Minister for Digital Transformation: : As is common with Materials 1 and 2, there is a concern that the operation in which humans see the risk of generative AI becomes enormous and the operation becomes difficult. When "GENAI" is customized and developed for the Cabinet Office and each Ministry, it may be necessary to increase the granularity of the description of the guidelines in accordance with the actual state of use by each Cabinet Office and each Ministry. It is also considered that it will be a burden for humans to comply with all the rules of the guidelines in the future. For example, regarding the matter of "Information designated" confidential "shall not be entered in prompts" in the Utilization rule template, it is necessary to consider implementing prompt regulations on the system from the beginning. Digital Agency needs to take progressive measures.
TORISAWA: In the future, it may be possible to automatically create the summary of the periodic report in Document 1. I think it would be good to consider such a direction.
Chairman: Kadobayashi: In order to improve the explainability and readability, based on the discussion at the plenary session, it would be good to consider it in the revision in the next fiscal year or later.

Chai Shan Members: Regarding the procurement check sheet and the contract check sheet, we would appreciate it if the contents of the comments made beforehand were also reflected. The content of the guidelines is also enormous, and it is considered that checking work on site is also difficult. Since we believe that past precedents are the most helpful in such checking situations, it would be good if we could further improve the horizontal development of use cases and the information sharing system.

Chairman: Kadobayashi: This concludes the exchange of views. If there are any points that need to be revised following today's discussions, I would like to ask I, as Chairman, to be responsible for making revisions while consulting with Digital Agency and, as necessary, individual members. If there is a need to make revisions, I will make the revisions myself, and I will proceed by reporting the details of the revisions to you all.

In the closing session, Mr. Minister for Digital Transformation stated that he would like to revise the guidelines and promote the utilization of "GENAI" by each ministry and agency at full speed for the purpose of making Japan the easiest country in the world to utilize AI and for the purpose of balanced AI utilization by the administration.