Ubie, as JaDHA Working Leader, Collaborates with AI Safety Institute (AISI) to Formulate 'AI Safety Evaluation Perspective Guide in the Healthcare Sector'

Japan Digital Health Alliance (hereinafter JaDHA), in which Ubie, Inc. (Headquarters: Chuo-ku, Tokyo; Co-CEOs: Yoshinori Abe and Kota Kubo; hereinafter 'Ubie') participates as a working leader company with the mission 'To guide people to appropriate medical care with technology', has formulated the 'AI Safety Evaluation Perspective Guide in the Healthcare Sector' (hereinafter 'this guide') in collaboration with the AI Safety Institute (hereinafter 'AISI') to accelerate the safe social implementation of generative AI technologies, including Large Language Models (LLMs), in the healthcare sector.

This guide refers to the evaluation perspectives formulated by AISI and presents specific evaluation methods for practitioners that reflect the sensitivity and risks unique to medical and healthcare fields.

AISI: Press Release https://www.ipa.go.jp/pressrelease/2026/press20260403.html

Background and Purpose In recent years, generative AI has brought 'once-in-a-decade innovation' to the healthcare sector, such as streamlining physicians' tasks and supporting patient communication. On the other hand, there are many challenges unique to the healthcare sector, including the risk of health damage caused by hallucinations (generation of false information), advanced privacy protection, and ensuring security.

As discussed at the 'Hiroshima Global Forum for Trustworthy AI' held in January 2026, building 'Trustworthy AI' has become a top priority for the international community.

Ubie, as the working leader of JaDHA's Healthcare SWG, led the formulation of this guide. We have aimed to create an environment where businesses can ensure safety from the development and design stages, balancing business value with peace of mind and safety.

Main Features of the Guide This guide has the following features so that even companies with few experts can easily utilize it.

**1. Setting Evaluation Methods at 5 Development and Design Stages along the AI Lifecycle** - Product Design: Clarifying the product's purpose and use cases, risk assessment, and building a governance structure. - Model Selection: Selecting a model suitable for the application and evaluating its safety. - Product Implementation: System architecture, prompt design, and guardrail implementation. - Product Verification: Comprehensive testing/verification and risk assessment. - Product Introduction and Operation: Monitoring and continuous improvement in the production environment.

**2. 10 Multifaceted Evaluation Perspectives and Specific Risk Scenarios for Each** - Output Control of Harmful Information: The risk that dangerous medical/health information (promoting self-harm/violence, baseless treatments, etc.) is output, causing direct harm to patients' lives/health or medical professionals' work. - Prevention of False/Misleading Information Output/Induction: The risk that hallucinations generate fictitious evidence or incorrect drug information, causing direct harm to patients' lives/health or medical professionals' work. - Fairness and Inclusivity: The risk that AI accuracy and quality degrade for patients with specific attributes (age, gender, race, region, etc.), resulting in disadvantages. - Dealing with Uses where High Risk is a Concern (High-Risk Use/Use for Unintended Purposes): The risk of regulatory violations occurring due to 'unintended use' where non-SaMD (Software as a Medical Device) is used as a de facto medical device. - Privacy Protection: The risk that medical/health information, including sensitive personal information, is leaked or misused, violating patients' privacy. - Ensuring Security: The risk that attacks such as prompt injection cause the tampering of medical information or the leakage of confidential data. - Explainability: The risk that AI outputs lack transparent reasoning, leading to incorrect actions by medical professionals or patient distrust. - Robustness: The risk that output quality becomes unstable against diverse inputs like dialects, abbreviations, and non-standard medical terms, leading to incorrect judgments. - Data Quality: The risk that outputs based on inaccurate or outdated medical data cause direct harm to patients' lives/health or medical professionals' work. - Verifiability: The risk that it is difficult to conduct post-verification or third-party audits, making it impossible to determine the cause of problems and damaging social trust.

AI Safety Evaluation Perspective Guide in the Healthcare Sector This guide can be downloaded from the following URL. https://aisi.go.jp/output/output_information/260402/

Main Target Audience - Executives and Business Managers - Product Managers (PM) - Engineers (Development), ML Engineers/Data Scientists, QA Engineers/Testers - UX Designers - Medical Experts/Domain Experts - Legal and Compliance - Security Personnel

Review Structure - Ubie, Inc. (SWG Leader) - Awarefy Inc. - Ajinomoto Co., Inc. - SB Intuitions Corp. - CMIC HOLDINGS Co., Ltd. - SherLOCK Inc. - Japan Digital Health Alliance (JaDHA) - MICIN, Inc. - The Tokyo Foundation for Policy Research - Mitsubishi Research Institute, Inc. (SWG Secretariat)

Future Outlook As a JaDHA working leader, Ubie positions this guide as a practical handbook for realizing trustworthy AI and will update it as appropriate in response to rapidly changing generative AI technologies, social conditions, and international regulatory trends. Through this, we will contribute to the safe social implementation of AI in the healthcare sector and the creation of sustainable business value. Furthermore, to enable this guide to be loaded and utilized by AI services, a portion of the guide will be published in markdown format. We also plan to release a practical guide featuring examples of prompts and agent skills in the future.

Mayu Inoue, Ubie, Inc., Representative of Accelerator Division / Counselor of Policy and External Affairs, JaDHA WG4 Leader Following rule-making activities for generative AI at the Japan Digital Health Alliance (JaDHA) from 2024, we took on the challenge of creating the 'AI Safety Evaluation Perspective Guide in the Healthcare Sector' from this fiscal year in collaboration between AISI and JaDHA. I would like to take this opportunity to express my deepest gratitude to the companies and stakeholders who cooperated with us. Ensuring safety and reliability is often considered a trade-off with promoting innovation, but in today's 'With AI' era, I believe that implementing 'Trustworthy AI' is exactly the engine for business growth and outcome creation. We will continue to work together as an industry to balance the promotion of innovation and the development of a safe and secure environment in the healthcare sector.

Masahiro Kazama, Ubie, Inc., Chief AI Officer (CAIO) This guide was created so that it can be utilized as a practical handbook when developing generative AI products considering AI safety in the healthcare sector. I hope it will help more products to be safely and securely implemented in society in the healthcare sector in the future. We also plan to publish examples of utilizing prompts and agent skills to organize AI safety perspectives specialized for each company's use cases. We will continue to work toward the realization of 'Trustworthy AI' in the healthcare sector. Finally, I would like to express my sincere gratitude to the companies, academia, and stakeholders who cooperated in creating this guide.

Related Information Press Release: JaDHA and AI Safety Institute (AISI) Collaboration Project begins discussing the development of a framework for AI safety evaluation in the healthcare field (July 9, 2025) https://jadha.jp/news/news20250709.html

Press Release: JaDHA announces the first-half activity results of the AISI Project Demonstration WG Healthcare SWG in collaboration with the AI Safety Institute (AISI) (October 2, 2025) https://prtimes.jp/main/html/rd/p/000000183.000048083.html

About Ubie, Inc. Founded in May 2017 by a medical doctor and an engineer, Ubie is a health-tech startup with the mission 'To guide people to appropriate medical care with technology'. With AI as its core technology, it develops and provides 'Ubie', which guides consumers to appropriate medical care, and 'Ubie Medical Navi', a service package for medical institutions that supports improving the quality of medical care. We are promoting the creation of a society where everyone can access medical care suited to them.

Location: Nihonbashi Life Science Building 4 5F, 3-8-4 Nihonbashi Honcho, Chuo-ku, Tokyo 103-0023 Established: May 2017 Representatives: Co-CEO & Medical Doctor Yoshinori Abe, Co-CEO Kota Kubo URL: https://ubie.life