Medical AI startup Aillis, Inc. announced the integration of Sakana AI’s Japanese-specialized LLM into its physician-targeted evidence search tool, “Evidence Finder.” This move is driven by government policies promoting Sovereign AI and a pressing demand to enhance AI accuracy and citation reliability in clinical settings.

Key Announcement Points and Background of the Agreement

Aillis, Inc., a developer of AI-powered medical devices, announced on September 14, 2026, an agreement to integrate a Japanese-specialized Large Language Model (LLM) developed by Sakana AI Co., Ltd. into its clinical evidence search tool, “Evidence Finder.” Founded in 2017 by practicing physician Sho Okiyama and his team, Aillis is recognized as a pioneer in medical DX through industry-government-academia-medical collaboration, having developed and commercialized the influenza-detection AI medical device “nodoca.” This partnership fuses Aillis’s proprietary search and verification algorithms with Sakana AI’s cutting-edge Japanese LLM technology. Japan’s medical administration and national policy frameworks—such as the Grand Design for Economic and Fiscal Management and Reform 2026 (Honefutoshi Policy 2026)—explicitly advocate for the adoption of “Sovereign AI” (domestic AI infrastructure). Consequently, this technical integration is drawing significant attention as a pioneering real-world clinical implementation aligned with national strategy. The image below illustrates an overview of this partnership.

Features and Capabilities of the Target Platform: Evidence Finder

Evidence Finder, the platform incorporating Sakana AI’s model, is a dedicated tool engineered to help clinical physicians swiftly retrieve and analyze medical evidence. Traditional general-purpose AI chat tools face critical hurdles in healthcare, notably hallucinations—outputting plausible yet false information—and an inability to verify source literature. Evidence Finder addresses these challenges through the following core capabilities:

• Automatically searching medical literature databases (such as PubMed) in response to clinical inquiries and explicitly citing source documents
• Featuring a built-in standard “Verify” function that automatically validates the actual existence of cited literature
• Providing a streamlined clinical workflow for physicians to rapidly access primary sources

While Evidence Finder is not classified as a medical device intended for direct diagnosis or treatment of individual patients, it serves as a powerful decision-support tool that significantly elevates the precision and efficiency of clinical information retrieval. Combining Aillis’s patent-pending algorithms with Sakana AI’s core capabilities has further reinforced the reliability of medical search.

Technical Highlights and Benchmark Results of Sakana Namazu

Model Performance Achieving 96.4% on the National Medical Practitioners Qualifying Examination

In February 2026, an evaluation model of Evidence Finder powered by “Sakana Namazu”—Sakana AI’s Japanese-specialized LLM—was benchmarked against the 120th National Medical Practitioners Qualifying Examination. The model achieved an outstanding score of 482 out of 500 points, yielding a 96.4% accuracy rate alongside a 100% answer completion rate. As of September 1, 2026, this stands as the highest recorded score among domestic foundation models specified under the Ministry of Economy, Trade and Industry’s GENIAC (Generative AI Accelerator Network) initiative. To measure the model’s intrinsic textual comprehension and medical reasoning capabilities, the image-rejection restrictions required in the commercial release were temporarily disabled during testing. This result rigorously demonstrates advanced mastery of medical terminology and deep reasoning tailored to Japan’s unique clinical context. The graph below shows the performance trends.

Strategic Significance of Adopting the Domestic LLM “Sakana Namazu”

Adopting a domestic LLM (Sovereign AI) in healthcare holds profound strategic value that extends far beyond computational benchmarks. Medical data comprises sensitive personal health information and highly specialized clinical knowledge, giving rise to persistent concerns regarding data security and privacy risks associated with transmitting data to overseas servers. Key advantages of integrating Sakana Namazu include:

• Exceptional nuance handling for the Japanese language, domestic medical systems, and clinical chart terminology
• Guaranteed data sovereignty and high-level security by maintaining domestic data management
• Mitigation of operational risks linked to sudden specification changes or service disruptions of foreign models

Since its inception, Aillis has driven medical AI development through an all-Japan effort, receiving recommendations from the Japanese Association for Acute Medicine and leveraging the National Institute of Advanced Industrial Science and Technology’s (AIST) supercomputer ABCI. The adoption of Sakana Namazu serves as an exemplary model for balancing data security with technological self-reliance in Japan’s medical DX landscape.

Clinical Implementation and Future Outlook

Ensuring Safety and Clarifying Legal Responsibility in Medical AI

As generative AI applications accelerate in clinical settings, adherence to strict operational guidelines—such as the Ministry of Health, Labour and Welfare’s “Generative AI Guidelines (Medical Version)”—becomes paramount. These guidelines explicitly define generative AI as an auxiliary decision-support tool, affirming that ultimate legal and clinical responsibility for diagnosis and treatment rests solely with the attending physician. In the event of a medical accident, legal liabilities—including civil damages or criminal charges such as professional negligence resulting in injury or death—are evaluated based on whether the physician breached their duty of care, rather than held against the AI developer. By design, Aillis’s Evidence Finder does not automate diagnosis; instead, it renders literature evidence transparent, empowering physicians to make informed decisions under their own professional responsibility. This system architecture underscores the vital balance between leveraging AI convenience and implementing robust governance to prevent misdiagnosis.

Future Horizons and Key Focus Areas in Clinical AI Deployment

The future medical AI landscape is moving beyond standalone image-diagnostic tools or voice transcription software toward fully integrated clinical ecosystems. Having raised a cumulative total of 5 billion yen in funding, Aillis continues to expand the nationwide adoption of its influenza-detection AI “nodoca” while advancing algorithms for new therapeutic areas. Following this strategic collaboration with Sakana AI for Evidence Finder, key developments to watch include:

• Enhanced data integration with digital health infrastructure, including Electronic Health Records (EHR) and Personal Health Records (PHR)
• Seamless clinical workflows covering initial intake, evidence search, and automated medical charting support
• Establishment of co-creative medical infrastructure that mitigates regional specialist shortages and preserves clinical implicit knowledge

Reducing administrative burdens for physicians and granting them more time for direct patient interaction directly enhances healthcare quality and sustains medical service delivery. In a domain where advanced technology intersects with stringent regulation, the implementation framework pioneered by Aillis and Sakana AI represents a major catalyst for transforming real-world healthcare.

References

• [Aillis Integrates Sakana AI’s Japanese-Specialized LLM into Physician Tool “Evidence Finder”] https://aillis.jp/news/post/20260914
• [Evidence Finder Product Overview] https://overview.evidencefinder.jp/
• [Sakana Namazu Official Page] https://sakana.ai/namazu/
• [Aillis, Inc. Official Website] https://aillis.jp/

【#MedicalAI #Aillis #SakanaAI #EvidenceFinder #SakanaNamazu #SovereignAI #MedicalDX #GenerativeAI】