Guidelines for Ensuring the Proper Development and Use of AI-Related Technologies (Provisional Translation)

【Provisional translation】 ※Please refer to the original text for accuracy Guideline for Ensuring the Appropriateness of Research & Development and Utilization of Artificial Intelligence-Related Technology December 19, 2025 Decision of the Artificial Intelligence Strategic Headquarters 【Provisional translation】 Table of Contents 1. Basic approach to ensuring appropriateness in Japan (1) Positioning of this guideline (2) Concept of ensuring appropriateness in this guideline (3) Basic policy for ensuring appropriateness 2. Matters to be especially addressed by R&D institutes and utilization business operators (1) Ensuring overall appropriateness through AI governance (2) Ensuring transparency to build trust with stakeholders (3) Ensuring sufficient safety (4) Maintaining a safe environment through business continuity (5) Consideration for stakeholders based on the importance of data as the foundation of AI Innovation 3. Matters to be especially addressed by national and local governments (1) Promoting innovation through active and leading utilization of AI (2) Improving AI literacy throughout the entire society (3) Examining appropriate approach to AI governance (4) Fulfilling accountability as an administration 4. Matters to be especially addressed by citizens (1) Responsible use of AI based on principles of human-centric AI (2) Appropriate use based on AI Literacy 【Provisional translation】 1. Basic Approach to Ensuring Appropriateness in Japan (1) Positioning of this guideline This guideline, based on Article 13 of the Act on Promotion of Research & Development(R&D), and Utilization of Artificial Intelligence-Related Technology (AI Act, Act No. 53 of 2025), is formulated in accordance with international norms toward realization of trustworthy AI , aiming to encourage voluntary and proactive efforts by all stakeholders 1 —including businesses, citizens, etc. —for the appropriate implementation of R&D and utilization of AI, The structure of this guideline is as follows: Section 1 presents the main elements and basic policies necessary for ensuring appropriateness in AI R&D and utilization for all stakeholders. Sections 2 and onward describe specific matters that each stakeholder should especially address, based on Section 1. All stakeholders are invit ed to recognize and understand the main elements necessary for ensuring appropriateness. Furthermore, regarding the matters to be addressed, stakeholders are required to respond appropriately at a suitable level, taking into account their scale, position, and the risks posed by AI, as well as the technologies and knowledge available at the time. Japan will develop a framework centered on this guideline as an international model for development, utilization and spread of trustworthy AI as well as promote international cooperation on construction of AI governance , continuously leading global discussions, based on achievement 2 of leading the “Hiroshima AI Process,” which is a framework to make an international rule regarding AI. (2) Concept of ensuring appropriateness in this guideline AI contributes to economic growth and the advancement of peopleʼs lives and it is important to promote its social implementation and innovation. However, AI also poses various risks: technical risks such as misjudgment and hallucination , 3 social risks such as the generation and spread of disinformation or misinformation, aggravation of bias or discrimination, use on crime , excessive 1 The term refers to the national government, local governments, research & development institut es, utilization business operators, and citizens, whose responsibilities are stipulated in Articles 4 through 8 of the AI Act. 2 At the G7 Hiroshima Summit in May 2023, the “Hiroshima AI Process” was launched. As an outcome under Japan's G7 Presidency, the “Hiroshima AI Process Comprehensive Policy Framework” concerning the development and use of advanced AI systems were compiled. The “Hiroshima AI Process Friends Group,” a voluntary framework of countries and regions that support the spirit of the Hiroshima AI Process, was established in May 2024, with 60 countries and regions participating as of December 2025. Also, in February 2025, the “Reporting Framework” commenced official operations, and as of December 2025, 24 organizations have submitted responses. 3 Generative AI refers to the phenomenon where it outputs information that differs from the facts in a plausible manner. 【Provisional translation】 dependence, infringement of privacy or property rights, increased environmental burden, employment or economic instability, and national security risks such as cyberattacks. These risks may change with technological progress of AI, and unknown risks may emerge, and social tolerance levels for these risks may also change. Therefore, in ensuring the appropriateness , this guideline does not provide a single definition or standard of appropriateness . Instead, under the expectation that each stakeholder will voluntarily advance its own initiatives based on the characteristics, intended uses, and purpose of the AI it researches, develops and utilizes, as well as its position and social role, and based on the principles set forth in “Social Principles of Human -Centric AI” (decided by the Integrated Innovation Strategy Promotion Council, March 29, 2019), this guideline identifies the main elements that should be considered as follows. Main Elements to Consider • Human-centricity  Respecting human dignity and fundamental human rights, and complying with the laws and regulations  Respecting diversity and inclusion ensures that everyone can benefit from AI, enabling people to pursue happiness and inclusive growth  The scope and conditions for u tilizing AI should be subject to a final decision made by a human • Fairness  Preventing and avoiding unjustified bias or discrimination in society resulting from AI utilization4 • Safety  Ensuring that AI utilization does not cause harm to human life, body, or property, etc.5 • Transparency 4 This includes ensuring that fairness is not undermined by biases, gender gaps and information manipulation that may arise from the use of AI. 5 This includes the freedom and honor that may be harmed by threats or defamation using deepfake technology to create fake videos, sexually altered images, or voice impersonations of others. 【Provisional translation】  Appropriately ensur ing transparency by disclosing information and securing post -hoc verifiability within the limits of what is necessary and technically feasible, to enhance reliability in AI6 • Accountability7  Fulfilling accountability within reasonable limits from technical, institutional, and societal perspectives, by clarifying responsibilities and establishing mechanisms to fulfill obligations, in light of the social impact of AI • Security  Appropriately ensuring AI security to reduce the risks such as unanticipated AI behavior or shutdowns caused by malicious manipulation • Privacy and personal information  Respecting and appropriately protecting privacy according to the importance of the data handled while complying with the Act on the Protection of Personal Information and other relevant laws and regulations • Fair competition  Even when resources related to AI are concentrated among specific entities, ensuring that unfair practices, including the improper collection of data leveraging their advantageous positions do not occur, thereby contributing to the promotion of fair competition • AI literacy  Recognizing that the socially acceptable level of risk posed by AI may change, acquiring the knowledge and capabilities to maximize the benefits and minimize the risks, while maintaining ethical awareness • Innovation  Striving to contribute to promote innovation while e nsuring sustainability including reducing environmental impact  Engaging in AI technology development that contributes to addressing social challenges  Addressing factors that hinder the utilization of AI 6 It is also important to advance the understanding of AI behavior and the process of generating output from input, thereby deepening our comprehension of the algorithms that contribute to AI output. 7 It means that individuals and organizations take responsibility for their actions and decisions and take action to fulfill th at responsibility. 【Provisional translation】 (3) Basic policy for ensuring appropriateness Based on the concept described in (2), the basic policies to be pursued to ensure appropriateness are outlined below: • Risk-based approach Identifying and evaluat ing the risks posed by AI, and implementing appropriate measures according to the potential impact based on the fields and purposes for which AI is utilized8 • Active involvement of stakeholders Entities affected by the benefits, risks, and other impacts of AI (hereinafter referred to as “stakeholders”) 9 shall actively participate in AI governance, and collabor ate with other stakeholders to address challenges. • Establishing a life-cycle AI governance framework Managing the risks posed by AI at a level acceptable to stakeholders as well as establishing AI governance that holistically addresses each stage of AI life-cycle—from research and development to societal implementation—to maximize the benefits • Agile Response Given the rapid pace of AI technological advancement and its insufficient predictability and explainability, enhancing the maturity of AI governance by responding flexibly and swiftly (hereinafter referred to as “agile”) to potential risks through the PDCA (Plan-Do-Check-Act) cycle 8 It is desirable to adopt diverse internal testing methods and independent external testing methods by combining various techniques such as red teaming, and to implement appropriate measures to address identified risks and vulnerabilities. 9 For example, this may include holders with data for the foundation of AI innovation, those who handle AI outputs and stakeholders who are affected by AI utilization but are not directly involved. 【Provisional translation】 2. Matters to be Especially Addressed by R&D Institutes and Utilization Business Operators Utilization business operators10 who develop and provide products and services utilizing AI should, considering that the AI they develop and provide may affect many stakeholders, utilize international norms11, international standards12, and other domestic guidelines13 related to R&D and utilization of AI, and especially address the following mat ters regarding the main elements necessary for ensuring appropriateness as indicated in Section 1(2). Furthermore, when providing the developed AI to third parties, AI R&D institutes14 shall address the following matters in particular regarding the main elements necessary for ensuring appropriateness as indicated in Section1(2). (1) Ensuring overall appropriateness through AI governance Establishing 15 operating, and continuously improv ing 16 AI governance —including organizational processes for identifying, assessing, and addressing risks throughout the entire AI lifecycle (design, development, provision, implementation, etc.), such as mechanisms for monitoring and evaluation involving managem ent, appropriate disclosure of related information, and implementation of 10 It refers to the utilization business operator as defined in Article 7 of the AI Act, including overseas business operators. 11 For example: Hiroshima Process International Code of Conduct for Organizations Developing Advanced AI Systems, and Hiroshima AI Process International Guiding Principles for All AI Actors. 12 For example: AI management system (ISO/IEC 42001). 13 It Refers to the Cabinet Office website (to be updated with domestic and international AI standards, guideline, etc.). 14 This refers to research and development institutes as defined in Article 6 of the AI Act. Regarding universities among such research and development institutes, consideration shall be given to respecting the autonomy of researchers and other characteristics of university research, as stipulated in Article 6, Paragraph 2 of the AI Act. 15 It is useful not only to build AI governance from scratch b ut also to leverage governance processe s already a pplied to existing IT systems and other areas. 16 At this juncture, to build a highly reliable organization, including measures to avoid risks related to AI and how to respond if risks become apparent , i t i s c o n s i d e r e d p o s s i b l e t o m a n a g e a n d c o n t r o l A I-related risks and fulfill social and ethical responsibilities by utilizing the “Hiroshima AI Process” Reporting Framework and establishing and operating a management system based on international standards (such as the AI management system (ISO/IEC 42001)). Proactively disclosing and explaining these initiatives is expected to enhance corporate value and secure competitive advantage. 【Provisional translation】 education and training, and managing AI risks at an acceptable level while maximizing the benefits AI brings. (2) Ensuring transparency to build trust with stakeholders Ensuring explainability within reasonable limits to building trust with stakeholders 17 regarding the origin of training data and generated outputs, including the appropriate implementation of intellectual property and privacy protections. Furthermore, when providing AI, supply ing users with information enabling its proper use (such as the AI's mechanisms and limitations, prohibited actions, data collection policies for training, and warnings regarding output reliability 18). (3) Ensuring sufficient safety Identifying and evaluating the risks of illegal activities such as various crimes including cyberattacks and fraud perpetrated through the misuse of AI, and implementing appropriate measures. Also, utiliz ing the latest technologies and knowledge to address and improve issues to suppress inappropriate outputs by AI such as hallucination, expansion of bias or discrimination, spread of misinformation or disinformation (including fake videos by deepfake technology and sexual altered images), and to prevent unintended operations or malfunctions of AI. In particular, given that the spread of misinformation and disinformation generated by AI poses a serious risk, R&D institutes and utilization business operators will strive to develop technologies that can determine whether something is generated by AI (digital watermarks, provenance management, APIs19, etc.) and will implement them as necessary. 17 To ensure appropriate transparency of training data, R&D institut es and utilization business operators will display the information (such as websites) used as the basis for AI outputs. When disclosure of training data is requested, it is desirable to respond as much as possible. Even in cases where technical constraints make it difficult to identify the relationship between AI outputs and training data, or when the requested training data constitutes a trade secret, it is expected that the matter will first be sincerely considered and discussed. 18 This includes warnings to prevent inappropriate user behavior that could lead to unfair bias or discrimination in hiring, performance evaluations, etc., or the spread of misinformation and dis information, as well as contact points and contact information for handling user inquiries. 19 Application Programming Interface: a mechanism for linking different applications (software) and systems to enable communication and data exchange. 【Provisional translation】 (4) Maintaining a safe environment through business continuity Operators of AI-based systems and service providers shall establish in advance a business continuity plan which shall define activities to be performed during normal operations, as well as methods and means for business continuity during emergencies, to minimize damage and enable early recovery in the event of system failures. (5) Consideration for stakeholders based on the importance of data as the foundation of AI innovation For AI innovation, it is important to secure high- quality data and use it appropriately. Based on this, to realize a virtuous cycle in which new creative activities are promoted by enriching high-quality data and developing an d providing reliable AI, businesses developing and providing AI will strive to continuously communicate with stakeholders, such as data holders of intellectual property, about the state of appropriate utilization, depending on the data usage situation. Furthermore, in particular, businesses that develop and provide AI with significant social impact shall endeavor to consider and implement measures aimed at establishing an ecosystem for returning benefits to such as date holders of intellectual property, and creating an environment in which creative activities, etc. can be conducted with peace of mind. 【Provisional translation】 3. Matters to be Especially Addressed by National and Local Governments The national government should especially address the following matters regarding the main elements necessary for ensuring appropriateness as indicated in Section 1(2). Local governments should, considering the diversity of their environments and challenges, respond as necessary with particular attention to the following matters in particular , according to local circumstances, regarding the main elements necessary for ensuring a ppropriateness as indicated in Section 1(2). When developing and providing AI, the national and local government s also address the matters indicated in Section 2. (1) Promoting innovation through active and leading utilization of AI Recognizing that widely disseminating practical use cases and key considerations in national or local governments is effective for promoting AI adoption, national and local governments will proactively lead the way in advancing AI utilization. They provide development and demonstration opportunities through public procurement. (2) Improving AI literacy throughout the entire society The national and local governments are required to promote the enhancement of AI literacy throughout society so that all entities —including, of course, national and local government employees—can understand issues related to ethics, laws, human rights, safety, and so on, and act with an awareness of their responsibilities as users. To this end, the government will continuously monitor the latest technological trends and practical applications of AI, examine associated risks and response measures, and present a concept to encourage voluntary initiatives by stakeholders. Furthermore, to ensure appropriate AI R&D and utilization by businesses and citizens, the government will actively promote education and guidance, including provision of contents that teach the basic usage and precautions for generative AI, and support for working adults in acquiring generative AI skills and knowledge. (3) Examining appropriate approach to AI governance The government will closely monitor domestic and international trends in AI governance, continuously review the state of AI governance, and respon d accordingly. This guideline and other domestic guidelines related to R&D and utilization of AI will be reviewed continuously and in an agile manner 【Provisional translation】 to reflect societal changes driven by technological advances in AI. In doing so, other domestic guidelines related to R&D and utilization of AI shall be reviewed and revised as appropriate to be consistent with the intent of this guideline and are easily understandable for businesses, citizens, and others. Furthermore, to reduce barriers to AI adoption in various contexts, with regard to issues that can be anticipated or that may arise when utilizing AI, the government will organize the issues and ideas regarding the interpretation and application of where responsibility lies, and strive to clarify interpretations as much as possible based on precedents, etc. Additionally, as AI operates across borders, international governance is essential alongside domestic efforts, the government will take the lead in establishing AI governance while also considering the need to ensure interoperability. (4) Fulfilling accountability as an administration When utilizing AI in government administration, to ensure the reliability of government administration, the government will implement appropriate measures to address its risks that fully consider the required standards, and fulfill our accountability to the public by ensuring that the basis for decisions remains as clear as possible. Moreover, each ministry and agency shall appoint an AI governance officer.20 Local governments shall clearly designate officers responsible for the appropriate utilization of AI and risk management. 20 “The Guideline for Japanese Governmentsʼ Procurements and Utilizations of Generative AI for the Sake of Evolution and Innovation of Public Administration” (A ppr o v e d b y t h e C o u ncil f o r t h e P r o m o t io n o f a D ig it a l S o cie t y E x e cu t iv e Bo a r d Meeting on May 27, 2025) stipulate s appointing a Chief AI Officer (CAIO) as a person responsible for formulating and promoting policies for the utilization of generative AI, overseeing the utilization status and risk management within the entire organization. When utilizing AI other than generative AI, it is also desirable to clearly designate responsible personnel as necessary. 【Provisional translation】 4. Matters to be Especially Addressed by Citizens Citizens shall respond with particular attention to the following matters regarding the main elements necessary for ensuring appropriateness as indicated in Section 1(2). (1) Responsible use of AI based on principles of human-centric AI Citizens, as the primary users of AI, shall comply with laws and regulations, recognizing that AI use may lead to violations of laws or harmful actions. In addition, citizens shall strive to understand not only the convenience of AI but also issues concerning ethics, laws, human rights, and safety, acting with awareness as responsible users. (2) Appropriate use based on AI literacy Citizens s trive to correctly understand the characteristics and mechanisms of AI and proactively acquire AI literacy. Furthermore, when utilizing AI, citizens shall understand the source and accuracy of the information obtained, make decisions under human judgment and responsibility, and refrain from inappropriate actions aimed at unfair bias or discrimination, slander, and spread of disinformation and misinformation. Additionally, citizens shall utilize AI-generated outputs (text, images, audio, video, etc.) in socially and legally appropriate ways.