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 ".

2nd Advanced AI Utilization Advisory Board

Last Updated:

  • : January 13, 2026 (Tuesday) 12:00 to 14:00
      1. Outline of generative AI System Periodic Report
      2. Government AI initiatives
      3. Implementation Report of the First Working Group of the Advisory Board on Advanced AI
      4. Trends in generative AI in Japan and Other Countries
      5. Draft revised policies to enhance guidelines for the procurement and utilization of generative AI for the advancement and renovation of public administration

At the beginning, Parliamentary Vice-Minister for Digital Transformation Kawasaki spoke about the ideal form of "reliable AI" that the government aims to realize, Digital Agency's initiatives related to AI utilization, including the Government AI, and the main agenda of the second Advisory Board meeting. Following this, Vice-Minister for Digital Transformation, Chief Officer of Digital Agency Misumi stated that he expects the government to further promote AI utilization.

1. Ministry and Agency generative AI System Periodic Report Summary

Reported the results of analysis of the status of utilization of generative AI by each ministry and agency using Document 1.

The main questions and opinions of participants are as follows.

Nabatame: It is important for governance in government, it is important for the government to make this initiative a common asset and promote its cross-sectional deployment across all ministries and agencies. As a keyword for promoting its utilization, how should it be standardized? For example, there may be cases where an initiative is implemented by one ministry or agency but not by another. It is necessary to raise the level of the initiatives of each ministry and agency through support for accelerating the introduction and to enhance the initiative of the entire government. For this purpose, a mechanism that assumes an integrated role, such as a quarterback function, is necessary.

Tamaki Okada:

"C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "Although it is excluded from the determination logic for" "C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "and the management of access-controlled function, we would like to make it easier to convey the relationship to readers by adding a statement in the guidelines that there is a description in the" "Uniform Standards for cybersecurity Measures for Government Agencies, etc." "Regarding the points of concern specific to the generative AI and the management of access-controlled." The performance targets have been included from the second periodic report. Case sharing is one of the roles of the Advisory Board, and as results will be accumulated through future periodic reports, cases will continue to be shared with the Advisory Board. As you pointed out, the importance of sharing best practices through Gennai has already been partially implemented, as described in Document 2, p. 9. In the future, cases will be shared with each ministry and agency as reference materials.

2. Initiatives of the Government AI

And an overview of the generative AI AI (Gennai) were reported using Reference 2.

The main questions and opinions of participants are as follows.

Yoshinaga: In relation to IP and the "procurement check sheet," I would like to express my opinion on the "Draft Code of Principles (tentative name) on the Protection and Transparence of IP for the Appropriate Utilization of

"C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "Although it is excluded from the determination logic for" "C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "and the management of access-controlled function, we would like to make it easier to convey the relationship to readers by adding a statement in the guidelines that there is a description in the" "Uniform Standards for cybersecurity Measures for Government Agencies, etc." "Regarding the points of concern specific to the generative AI and the management of access-controlled." p. 9, but specific use cases have not been grasped. It is possible to judge by checking the prompt, but since there is no such need at present, it has not been confirmed.

Member: It is difficult for a human to look at a prompt and check how to use it, but it may be possible to automatically grasp the purpose of use by asking the LLM how to use the prompt.

"C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "Although it is excluded from the determination logic for" "C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "and the management of access-controlled function, we would like to make it easier to convey the relationship to readers by adding a statement in the guidelines that there is a description in the" "Uniform Standards for cybersecurity Measures for Government Agencies, etc." "Regarding the points of concern specific to the generative AI and the management of access-controlled." Other ministries and agencies are concerned about information leakage, and Digital Agency will consider it to the extent possible.

Nabatame: It is important for governance in

"C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "Although it is excluded from the determination logic for" "C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "and the management of access-controlled function, we would like to make it easier to convey the relationship to readers by adding a statement in the guidelines that there is a description in the" "Uniform Standards for cybersecurity Measures for Government Agencies, etc." "Regarding the points of concern specific to the generative AI and the management of access-controlled." As you pointed out, we would like to identify the barriers to utilization. Since there is a possibility that issues for the utilization of generative AI may become clear from cases where verification progressed but did not reach the stage of normal use, we would like to promote the horizontal dissemination of information to each ministry and agency while conducting interviews.

Chairman: Mr. generative AI be conducted by each ministry and agency, and will Digital Agency conduct a hearing?

"C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "Although it is excluded from the determination logic for" "C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "and the management of access-controlled function, we would like to make it easier to convey the relationship to readers by adding a statement in the guidelines that there is a description in the" "Uniform Standards for cybersecurity Measures for Government Agencies, etc." "Regarding the points of concern specific to the generative AI and the management of access-controlled." Government, we would like to develop open data such as white papers and create an environment in which analysis is possible by entering simple questions into a AI. Domain expertise held by each ministry and office is often not shared even among adjacent departments. While paying attention to authority management, we need to consider ways to make this information available through AI agents, etc. Since there may be some information that can be made open to a certain extent within each ministry and office, we would like to continue to devise ways to make it available.

Member: 's information, but I would like to ask whether Digital Agency or the relevant ministries and agencies will decide on how to respond to information disclosure requests from the public regarding "Gennai." What is the status of consideration at this point?

"C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "Although it is excluded from the determination logic for" "C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "and the management of access-controlled function, we would like to make it easier to convey the relationship to readers by adding a statement in the guidelines that there is a description in the" "Uniform Standards for cybersecurity Measures for Government Agencies, etc." "Regarding the points of concern specific to the generative AI and the management of access-controlled." Basically, it is assumed that each department that holds the data subject to the information disclosure request will handle it.

Member: It is desirable that the government ministries and agencies be able to apply information disclosure requests in a unified manner. I would like Digital Agency to do its utmost.

Naganuma, Member: In December 2025, we conducted interviews with Keidanren-related companies regarding the issues presented at the first Advisory Board meeting to revise the Regarding page 10 of Document 2, it is desirable to have a monitoring mechanism that can understand how the AI is being used when the dashboard of the current utilization status of Gennai is expanded to other ministries and agencies in the future, and it is important to introduce a mechanism for verifying the effects along with a large-scale demonstration.

Tamaki Okada:

"C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "Although it is excluded from the determination logic for" "C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "and the management of access-controlled function, we would like to make it easier to convey the relationship to readers by adding a statement in the guidelines that there is a description in the" "Uniform Standards for cybersecurity Measures for Government Agencies, etc." "Regarding the points of concern specific to the generative AI and the management of access-controlled." will establish a CAIO to strengthen governance. Regarding the guidelines, training materials are provided from Digital Agency to each ministry and agency. Regarding common training for ministries and agencies, in addition to basic training, training for Gennai is also planned. Regarding the use of Gennai, as of August 2025, about half of Digital Agency employees had never entered a prompt. In light of this situation, we believe it is important to first promote the use of Gennai by enhancing training for beginners, and then prepare phased training for intermediate and advanced users.

Tamaki Okada:

"C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "Although it is excluded from the determination logic for" "C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "and the management of access-controlled function, we would like to make it easier to convey the relationship to readers by adding a statement in the guidelines that there is a description in the" "Uniform Standards for cybersecurity Measures for Government Agencies, etc." "Regarding the points of concern specific to the generative AI and the management of access-controlled." As you pointed out, the use of the output results is ultimately the responsibility of the staff who use generative AI, but the GOJ will remind them that it is necessary to take into consideration the historical background and values of the period in which the official document was prepared and to fully confirm the content.

Chairman: Mr. This issue is related to R & D, and it is desirable that it also be discussed at the Information-Technology Promotion Agency, Japan (AI) and the AI Safety Institute (IPASI), taking into account trends in other countries.

"Intellectual Property Rights, etc. Measures Reference Sheet" by could be summarized by Digital Agency and the Advisory Board and shared as alerts. If the number of users increases, efforts to improve risk training and literacy will become even more important in the future. At present, expanding the number of users itself is a challenge, but measures that look ahead to after the expansion are also necessary.

"C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "Although it is excluded from the determination logic for" "C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "and the management of access-controlled function, we would like to make it easier to convey the relationship to readers by adding a statement in the guidelines that there is a description in the" "Uniform Standards for cybersecurity Measures for Government Agencies, etc." "Regarding the points of concern specific to the generative AI and the management of access-controlled." Since one of the roles of the Advisory Board is to grasp risk case information and provide advice to prevent recurrence, if a risk case occurs, it will be shared with the Advisory Board and consideration will be given to measures to prevent recurrence. We will also consider training.

"3. Implementation report of the first working group

The secretariat reported the outline of the first working group meeting in Document 3.

The main questions and opinions of participants are as follows.

Yoshinaga: In relation to IP and the "procurement check sheet," I would like to express my opinion on the "Draft Code of Principles (tentative name) on the Protection and Transparence of IP for the Appropriate Utilization of , "Issue 1 Perspective of managing and reducing risks when the general public widely uses generative AI chatbots" Regarding the "Opinion on Al Harthi Nation Risk", I think a protocol that guides people to manned support is very important. At the end of 2025, I participated in an external event called the "AI Agent Expo". Since some of the erroneous responses by AI are caused by insufficient communication on the human side, AI agent products that assist communication for people who lack language have appeared. For example, a AI agent that responds to a prompt input by a human with only one word, such as "tax return," by supplementing the context while listening to the needs has appeared. Although the need for manned support may decrease in the future due to technological development, a protocol that guides people to manned support is necessary until the technology reaches a certain level.

Member: I fully agree with the problem of communication and hallucination. When classifying hallucination, there are two types of cases: cases where incorrect information is presented and cases of so-called "oversight" where relevant information exists but is not mentioned. Personally, when I needed to report the details related to the sale of real estate on my tax return in the past, I could not fully understand many special provisions, and as a result, I had an experience of asking a certified public tax accountant. Since there are many special provisions in the tax field, it is necessary to consider the risk that an AI will "overlook" them. In addition, in cases where an employee is assigned to respond after communication with generative AI, the content of the response on the employee side is considered to change compared to cases where communication with the AI is not involved, and it is important to have a mechanism to share information on issues that users tend to overlook in advance with the corresponding employee.

"C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "Although it is excluded from the determination logic for" "C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "and the management of access-controlled function, we would like to make it easier to convey the relationship to readers by adding a statement in the guidelines that there is a description in the" "Uniform Standards for cybersecurity Measures for Government Agencies, etc." "Regarding the points of concern specific to the generative AI and the management of access-controlled." Working Group implementation report will be shared with National Tax Agency.

Naganuma, Member: In December 2025, we conducted interviews with Keidanren-related companies regarding the issues presented at the first Advisory Board meeting to revise the We would like the Advisory Board to continue to share information on the areas in which the generative AI Rules are regularly revised, which are indicated as future directions in "Issue 1: Perspectives on managing and reducing risks when allowing the general public to widely use laws and regulations chatbots," because we believe these are particularly high-risk perspectives.

"C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "Although it is excluded from the determination logic for" "C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "and the management of access-controlled function, we would like to make it easier to convey the relationship to readers by adding a statement in the guidelines that there is a description in the" "Uniform Standards for cybersecurity Measures for Government Agencies, etc." "Regarding the points of concern specific to the generative AI and the management of access-controlled." National Tax Agency is a long-term project that is expected to be introduced in 2028. In the future, information will be shared with the Advisory Board as appropriate when consulted by the Working Group according to progress.

4. Trends related to generative AI in Japan and other countries

And generative AI in Japan and other countries are reported by the secretariat using Document 4.

The main questions and opinions of participants are as follows.

Kitamura, Use cases and technological trends in the private sector were introduced. It is essential for the Japanese government to strengthen strategically weak tasks and further develop strong tasks, as well as create new tasks for catch-up and understand the corresponding use cases and trends. Therefore, it is necessary to strategically consider the granularity and targets of the survey in the future.

Kitamura, A list of current AI-related guidelines from each Ministry and Agency was introduced. There are some guidelines that need to be translated into English in a timely manner from the perspective of enhancing the Japanese presence. As past experience, we have translated the NIST AI Risk Management Framework into Japanese in cooperation with related organizations in the United States. However, there were many difficulties in the translation process. In the future, as part of the government AI initiatives, it is possible that translated RAGs will be established as one of the RAGs in the generative AI User Environment (Gennai). However, at present, there is room for considering the establishment of translated dictionaries for AI-related guidelines from each Ministry and Agency in cooperation with related organizations.

"C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "Although it is excluded from the determination logic for" "C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "and the management of access-controlled function, we would like to make it easier to convey the relationship to readers by adding a statement in the guidelines that there is a description in the" "Uniform Standards for cybersecurity Measures for Government Agencies, etc." "Regarding the points of concern specific to the generative AI and the management of access-controlled." " Secretariat: When translating public documents such as the AI Guideline for Business Guidelines into English, the Government of Japan would like to consider following the same terminology, such as "Okinawa," so that there will be no discrepancies in terminology within the Government.

Nabatame: It is important for governance in to be tackled across governments. With regard to the comparison of government AI procurement and use rules in other countries, it is effective to examine Japan's measures while paying close attention to the trends of each country by sorting out both individual efforts and cross-sectional efforts by each government ministry and agency. The attitude of working together to solve issues will be an important message that shows the smooth relationship between Digital Agency and each government ministry and agency in the future.

5. Draft Revised Policies to Enhance Guidelines for Procurement and Utilization of generative AI for Evolving and InNovating Public Administration

5.

The main questions and opinions of participants are as follows.

Chairman: Mr. , a member of the Federation of Economic Organizations, will report on the results of the questionnaire.

Naganuma, Member: In December 2025, we conducted interviews with Keidanren-related companies regarding the issues presented at the first Advisory Board meeting to revise the Guidelines. There were no objections to the expansion of the scope of the guidelines, and many people said that it is necessary to expand the scope of the guidelines because there are concerns that a gap will arise between the private sector and the government if the scope is left fixed in fields with rapid technological progress. It was also pointed out that even if best practices have not yet been established, it is important to reflect what is considered to be the best at the time and to update them in an agile manner. There was support for the idea of applying the Common Principles in advance and expanding the scope gradually from high-risk AI. Regarding the high-risk criteria, the majority of people evaluated that the four axes were easy to understand. On the other hand, there were opinions that requested the presentation of specific examples of "operations where failure may have a significant impact." Regarding the "procurement check sheet" and "contract check sheet," there were opinions that requested a review and enhancement based on the latest technological trends and the development of use cases. In particular, regarding the provision of information on learning data, there were opinions that requested a flexible description method because it is difficult to disclose details in some cases from the perspective of confidentiality. In addition, from the perspective of reducing the burden on both the government and business operators, there were opinions that arranged the correspondence with external certifications such as ISO/IEC 42001 and ISO/IEC 27001 and requested the simplification or exemption of checks in cases where a certain level of certification has been acquired. It was pointed out that it is important to clarify the reference relationship with international documents such as the Hiroshima AI Process Reporting Framework from the perspective of ensuring interoperability. In addition, it was pointed out that it is important to have a mechanism to evaluate so-called white efforts, such as voluntary efforts to ensure clarity and reduce risks, not limited to reporting on accidents. In addition, regarding the current situation in which multiple guidelines are disclosed within the government, there were opinions that it is difficult to make cross-references because the nature and granularity of the documents differ, that priorities are difficult to understand, and that on-site implementation is difficult. Opinions were expressed that cross-references should be clarified and that consideration should be given to aggregation in the future.

Kitamura, . It is necessary to consider a balance so that the citizens do not take excessive responsibility for the generative AI system, which is highly black-box.

"C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "Although it is excluded from the determination logic for" "C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "and the management of access-controlled function, we would like to make it easier to convey the relationship to readers by adding a statement in the guidelines that there is a description in the" "Uniform Standards for cybersecurity Measures for Government Agencies, etc." "Regarding the points of concern specific to the generative AI and the management of access-controlled." has been reviewed by each member, but we will continue to pay attention to the creation of the Terms of Use of the Kansai System so as not to make the contents excessive for citizens.

"Intellectual Property Rights, etc. Measures Reference Sheet" by Regarding the revision of the contract check sheet, I think that the point of clearly stating that it is basically a quasi-mandate contract is an important message based on trends in the private sector.

"C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "Although it is excluded from the determination logic for" "C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "and the management of access-controlled function, we would like to make it easier to convey the relationship to readers by adding a statement in the guidelines that there is a description in the" "Uniform Standards for cybersecurity Measures for Government Agencies, etc." "Regarding the points of concern specific to the generative AI and the management of access-controlled." Based on the guidelines of the Japanese Deep Learning Association (JDLA), we would like to proceed with the materialization and present a revised draft at the next Advisory Board meeting.

"Intellectual Property Rights, etc. Measures Reference Sheet" by is currently under development. While it is a very important initiative, it is difficult to completely eliminate the risks related to copyrights, etc. It is not realistic to always check whether copyrights have been infringed on generative AI's outputs. Therefore, it is more effective to promote its use in low-risk areas than to focus on checking whether copyrights have been infringed.

"C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "Although it is excluded from the determination logic for" "C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "and the management of access-controlled function, we would like to make it easier to convey the relationship to readers by adding a statement in the guidelines that there is a description in the" "Uniform Standards for cybersecurity Measures for Government Agencies, etc." "Regarding the points of concern specific to the generative AI and the management of access-controlled." As a supplement, the" Reference Sheet for Intellectual Property Rights, etc. "is positioned as a reference material for clarifying the relationship with the existing guidelines and the" Procurement Check Sheet. "Important contents shall be reflected in the" Model of Rules for Use and Application of the generative AI System "and the" Procurement Check Sheet. "We would like to present the contents to the members of the Advisory Board at the next meeting.

Chairman: Mr. There have been cases of litigation involving intellectual property rights in large amounts overseas, so I would like you to respond while paying close attention to international trends.

Yoshinaga: In relation to IP and the "procurement check sheet," I would like to express my opinion on the "Draft Code of Principles (tentative name) on the Protection and Transparence of IP for the Appropriate Utilization of ," which is currently under consideration in Cabinet Office. I recognize that the purpose of the Act on Promotion of AI was to require companies to provide information on risks. However, the Code of Principles requires companies to disclose information on usage models and training data. While stating that "disclosure is not mandatory," the word "comply" in English expressions is very nuanced. Requiring private companies to disclose usage models and training data would be an excessive burden and would hinder the "counteroffensive" mentioned in the Basic Plan for AI. In the first place, the method of collecting information to create a AI model would fall under know-how for companies and could also fall under corporate secrets. Therefore, it is necessary to proceed with careful consideration. If it were included as a requirement in the procurement guidelines, it would limit the suppliers that can implement it, and there is a concern that it would hinder the entry of startups, etc., which is not desirable from the perspective of competition law. On the government side, there is a possibility that Gennai will be subject to information disclosure, which could pose information security risks. It is also necessary to pay attention to the fact that it could become a target of attacks by other countries and increase vulnerability. generative AI

"C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "Although it is excluded from the determination logic for" "C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "and the management of access-controlled function, we would like to make it easier to convey the relationship to readers by adding a statement in the guidelines that there is a description in the" "Uniform Standards for cybersecurity Measures for Government Agencies, etc." "Regarding the points of concern specific to the generative AI and the management of access-controlled." Principles Code, and will consider its response, taking into account the results of the public comments made by the Intellectual Property Office.

Chairman: Mr. AI - We have received negative opinions regarding the BOM (AI Bill of Materials) part. In particular, careful consideration is necessary regarding the disclosure of information on learning data. In the standardization field, there is a distinction between Normative Reference and Informative Reference. Sufficient consideration is necessary regarding whether or not to refer to the Principle Codes, and if so, whether to refer to the Principle Codes as Normative or as reference information. We hope to come up with ideas from February to March.

"C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "Although it is excluded from the determination logic for" "C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "and the management of access-controlled function, we would like to make it easier to convey the relationship to readers by adding a statement in the guidelines that there is a description in the" "Uniform Standards for cybersecurity Measures for Government Agencies, etc." "Regarding the points of concern specific to the generative AI and the management of access-controlled." As you pointed out, this guideline is a normative document, but its content includes informative elements. We will continue to consider the matter, including the possibility of selecting the best one according to the content.

Chairman: Mr. In international documents, there is a distinction between Annexes and Appendices. The meaning of Annexes and Appendices is often unclear in Japanese. It is desirable to clearly indicate whether they are normative or reference materials at the beginning of the document.

Kitamura, : International discussions on the AI BOM have started with the SBOM, but discussions on the extension of the AI system, the so-called AI BOM, have also been held. It is necessary to advance domestic discussions while taking into account such international trends. In addition, careful consideration should be given to the use of the term "compliance" in relation to the Principle Code. For example, in private sector, the term "compliance matrix" is sometimes used as a term with a strong meaning close to the essential conditions for bidding, and attention should be paid to its use in policy documents, etc.

Tamaki Okada: Regarding the risk assessment logic, I generally agree with the new logic. On the other hand, from the public's perspective, the deletion of "C. Whether classified information or personal information has been learned, etc.", which was the conventional assessment axis, may promote anxiety if the public who does not fully understand the national trend sees only this part. It is desirable to add a description in addition to the commentary rather than completely excluding it from the logic.

"C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "Although it is excluded from the determination logic for" "C. Presence or absence of Information designated" "Confidential" "or" "Private" "Information that requires confidentiality" "and the management of access-controlled function, we would like to make it easier to convey the relationship to readers by adding a statement in the guidelines that there is a description in the" "Uniform Standards for cybersecurity Measures for Government Agencies, etc." "Regarding the points of concern specific to the generative AI and the management of access-controlled."

Chairman: Mr. , I would like the Secretariat to take stock of your comments. This concludes the exchange of views.