AI Policy

AI is a tool that supports human dignity; it does not replace human judgment, responsibility, or dignity. The Office prioritizes the dignity of care recipients, their families, care workers, foreign staff, and all other stakeholders. Where efficiency gained through AI would compromise dignity, the Office refrains from using AI.

AI is designed and operated to support the equitable participation and benefit of all persons, regardless of gender, age, nationality, religion, language, sexual orientation, gender identity, disability, or employment status. The Office treats diversity not as a cost but as a value, and continuously verifies that AI does not exclude or discriminate against minorities.

Final responsibility for AI outputs always lies with human beings. The Office regards as inviolable the principle that "determination is done by code, articulation is done by AI, and final judgment is done by humans," and internalizes this principle as part of its organizational culture. Even where AI appears to make autonomous decisions, it is humans who choose to deploy it, and responsibility accordingly rests with humans.

This Policy applies to all of the following activities.

• All uses of AI (including generative AI, RAG, AI agents, and physical AI) by the representative, staff, or contractors of the Office in the course of business.
• AI functions embedded in the CAMP (Care AI Management Professional) system — including manga and video generation, training report generation, and AI chatbot features — in their planning, production, and provision.
• All chatbots and AI tools built by the Office through customization of Gemini GEM, NotebookLM, ChatGPT, Claude, or other foundation models.
• Services provided by the Office — including the CAMP certification program, training, and consulting — that recommend or instruct the use of AI.
• All deliverables produced with the aid of AI, including books, columns, lecture materials, videos, and manga published by the Office.

In light of the classification of actors set forth in the AI Guidelines for Business (Version 1.2), the Office positions itself as follows.

• AI Provider: The Office provides the CAMP system's mandatory training package, as well as various chatbots built by customizing Gemini GEM and NotebookLM.
• AI User: The Office routinely uses multiple foundation models (such as Claude, ChatGPT, and Gemini) for business execution, manuscript writing, document preparation, and CAMP manga/video production.
• AI Developer (Limited): Although the Office does not develop foundation models themselves, it recognizes that because it customizes knowledge files and system prompts, it may be regarded, in part, as equivalent to a "developer" under the AI Guidelines for Business.

AI is positioned as a tool that augments human capability and shall not infringe upon the rights guaranteed by the Constitution or internationally recognized human rights. In particular, the Office prioritizes consideration of impacts on persons who may be in vulnerable positions, including care recipients, foreign staff, and LGBTQ+ persons.

The Office conducts pre-deployment risk assessments to ensure that AI does not harm the life, body, property, mental well-being, or environment of any person involved. Particularly careful verification shall be conducted for applications in which safety is directly linked to human life, such as fall prediction and medication support in care settings.

The Office takes care to eliminate potential bias, and, while acknowledging that unavoidable bias exists, evaluates its impact from the standpoint of fairness that respects human rights and cultural diversity. Disparities in treatment among foreign staff based on country of origin or religion, and disparities in services to care recipients based on gender or family composition, are regularly reviewed.

The Office complies with the Act on the Protection of Personal Information and related laws and regulations. In particular, detailed input restriction rules (Article 18) are established with respect to sensitive personal information concerning care recipients, such as medical history, long-term care certification, family circumstances, and religion, as well as information concerning foreign staff, including residency status, country of origin, and religion.

The Office prevents unintended changes to, or stoppage of, AI behavior caused by unauthorized manipulation. The Office verifies, on each occasion, the terms of service of foundation model providers, whether inputs are used for training, and the location of data storage, and, as a general rule, uses enterprise-grade contracts (such as ChatGPT Enterprise, Claude for Work, and Gemini Workspace).

The Office provides information about AI itself and about AI systems and services to stakeholders to the extent that is reasonable and technically feasible. Where AI is used in columns, books, training materials, or CAMP content published by the Office, the fact of such use is clearly disclosed.

The Office fulfills its accountability obligations to relevant stakeholders to a reasonable extent with respect to information concerning data sources, the traceability of AI decision-making, and the status of responses to risks.

The Office provides AI ethics and risk training for the representative and all staff at least once per year. Through the CAMP certification program, the Office also provides opportunities for external stakeholders — including long-term care providers, administrative scriveners, tax accountants, and social insurance labor consultants — to enhance their AI literacy.

In business competition involving the use of AI, the Office complies with the Antimonopoly Act and related laws and regulations, and does not engage in improper information enclosure or the suggestion of AI-based price coordination.

Recognizing that the loss of AI utilization opportunities due to excessive caution in the long-term care sector is also a social loss, the Office actively promotes the use of AI under appropriate risk management.

In order to implement the risk-based approach required by the AI Guidelines for Business (Version 1.2), the Office classifies all AI-utilizing operations into the following three tiers. The higher the tier, the more stringent the controls.

Tier determination for each operation shall be made, in the first instance, by the staff member in charge of the operation, and finalized by the representative (Michihiro Kohama) or the person responsible for AI governance. Where there is doubt regarding the determination, the more stringent tier shall be applied.

The following information shall not be entered into external generative AI, except in enterprise-grade contract environments where use for training can be explicitly opted out of.

• Identifiable personal information of care recipients and their families, such as name, address, and contact information
• Sensitive personal information, such as medical history, long-term care level, disability status, medication information, and progression of dementia
• Residence card information, passport information, and details of the country of origin, religion, and family situation of foreign staff
• Non-public management, financial, and litigation-related information of client and partner organizations
• Tax return data, tax-related information, and payroll records of clients
• All information that is subject to a non-disclosure agreement (NDA) between legal entities
• Personally identifiable information contained in CAMP customer training data
• Credentials such as passwords, API keys, authentication tokens, and encryption keys

The prohibitions set forth in Article 18 may be exceptionally waived only upon written approval by the representative or the person responsible for AI governance, and only where all of the following conditions are met.

• A contract under which input data is not used for training (such as an enterprise contract or a Zero Data Retention agreement) has been concluded
• The explicit consent of the data subject, or a lawful basis under the Act on the Protection of Personal Information, exists
• Measures for anonymization or pseudonymization have been implemented
• Procedures for preserving input/output logs and for deletion after the purpose is achieved have been established

The Office is not directly involved in issues regarding the training of foundation models themselves. However, with reference to the Agency for Cultural Affairs' "Views on AI and Copyright" (published March 2024), the Office shall cease using any foundation model whose training data is determined to have serious issues.

When using AI to produce images, manga, videos, book manuscripts, CAMP training materials, article drafts, and similar works, the following verification flow is implemented.

• Verify through reverse-image search, similar-text search, and other means that the output does not rely upon or resemble existing copyrighted works.
• As a rule, avoid directly specifying the names of particular authors or works in prompts.
• Where third-party materials are included in training or prompts, verify the licensing conditions and commercial-use permissibility in advance.
• When using generated content for commercial purposes, verify, on each occasion, the terms of use of generation tools such as Nano Banana PRO.

When using AI to generate explanatory materials, training materials, or work instructions for foreign staff, attention is paid to the following matters.

• Because the accuracy of AI-based automatic translation is not perfect, important employment conditions, safety instructions, and contractual matters shall be finalized only after review by a human translator or a person proficient in both Japanese and the target language.
• The Office inspects AI outputs for inappropriate or biased content concerning religious and cultural considerations, such as Islamic prayer times, halal dining, and Christian Sunday worship.
• On the premise that AI may generate incorrect information about residency statuses such as the Training and Employment Program and the Specified Skilled Worker Program, cross-checking against official information from the Immigration Services Agency is mandatory.

When designing or recommending predictive care involving care recipients through AI (such as fall prediction, prediction of dementia progression, and prediction of unauthorized departure), the following matters are observed.

• Prediction results are treated as reference information and shall not be used as the sole basis for care plan changes or physical/behavioral restrictions.
• The Office explains to care recipients themselves and to their families the fact of AI use, the scope of such use, and the existence of the right to refuse.
• Care is taken so that AI predictions do not lead to the uniform or simplified treatment of the individuality, life history, or personality of care recipients.

With respect to recruitment support, staff assignment, and training material creation using AI, the Office conducts periodic bias verification to ensure that no discriminatory output occurs on the basis of gender, gender identity, sexual orientation, age, prior service in the Self-Defense Forces, or similar factors.

The Office establishes the following as the Golden Rule of AI utilization, and applies it to all operations.

Responsibility for judgments made by customers based on outputs from the CAMP system and AI tools (such as Pods) provided by the Office shall, as a rule, rest with the customer. However, in the event of serious defects in the system design, knowledge files, or prompt design prepared by the Office, the Office shall bear responsibility to a reasonable extent. This demarcation of responsibility is expressly stated in customer agreements and terms of service.

For the purposes of this Policy, "incident" means any of the following situations.

• A situation in which sensitive personal information has been leaked externally as a result of AI output
• A situation in which AI output indicates an erroneous interpretation of a regulatory framework, creating a risk of errors in a customer's addition calculations or long-term care benefit billing
• A situation in which AI output contains discriminatory or human-rights-infringing content, and the Office has received a complaint from an affected person or their family
• A situation in which AI-generated content is alleged to infringe the copyright or other intellectual property rights of a third party
• A situation in which inappropriate or erroneous training content has been distributed through the generative AI functions of the CAMP system

The response to an incident shall be as follows.

• The person who discovers the incident shall immediately report it to the representative or the person responsible for AI governance.
• The representative or the person responsible for AI governance shall temporarily suspend use of the AI in question and identify the scope of impact.
• As necessary, the Office shall notify the Personal Information Protection Commission, affected persons, and relevant administrative agencies.
• The Office shall conduct a root-cause analysis and formulate measures to prevent recurrence. This Policy shall be revised as necessary.

The Office provides training covering the following content for the representative and all staff at least once per year. Records of attendance are retained for three years.

• Overview and updates of the AI Guidelines for Business (Version 1.2)
• Each provision of this Policy (particularly tier classification and prohibited input information)
• Case studies of AI-related incidents (domestic and international)
• Specific examples of hallucination and methods for detecting them

Through the CAMP (Care AI Management Professional) certification program and mandatory training package, the Office provides external stakeholders — including long-term care providers, management consultants, tax accountants, social insurance labor consultants, and administrative scriveners — with opportunities to enhance their AI governance literacy. The Office positions this as part of its social responsibility.

This Policy shall be reviewed when any of the following events occurs. Even in the absence of such events, a regular review shall be conducted at least once per year.

• Revision of the AI Guidelines for Business (publication of Version 1.3 or later)
• Revision of Cabinet Orders or Basic Plans under the AI Promotion Act
• Amendment of the Act on the Protection of Personal Information or the Copyright Act
• The occurrence of a serious incident
• Major feature updates to the CAMP system
• Significant advances in AI technology, particularly in AI agents and physical AI

The Office accepts inquiries, complaints, indications of output errors, and claims of rights infringement concerning its AI activities and the CAMP system at the following contact point. As a rule, an initial response shall be provided within ten business days of receipt.

This Policy shall take effect on May 1, 2026.

In establishing this Policy, the following standards, guidelines, and templates were consulted.

• MIC & METI, "AI Guidelines for Business (Version 1.2)" (published March 31, 2026)
• Act on the Promotion of Research and Development and Utilization of Artificial Intelligence-Related Technologies (Act No. 53 of 2025)
• Japan Deep Learning Association (JDLA), "Guidelines for the Use of Generative AI" Version 1.1 and the Image Edition
• Agency for Cultural Affairs, "Views on AI and Copyright" (published March 2024)
• Risk assessment methodology under the Kobe City AI Ordinance (enacted 2024)
• Fujitsu Limited, "AI Ethical Impact Assessment" handbook and "Fujitsu AI Ethics Karuta"
• Structure of requirements in ISO/IEC 42001:2023 (AI Management System)