Reducing Concerns for Generative AI Adoption: Patent Application Filed for Next-Generation Secure AI Platform Fusing Secret Sharing (AONT) Technology and RAG

Aegis Application Inc. (Representative Director and President: Yasunori Kobayashi), ZenmuTech Inc. (Representative Director and CEO: Yasuhisa Abe), and Technica AI Inc. (Representative Director: Keiki Tsuyu) have jointly developed a "Secret Sharing x RAG (Retrieval-Augmented Generation) Solution" that enables secure AI utilization by transforming confidential information into meaningless, irrecoverable distributed fragments and storing them dispersedly. In this joint development, ZenmuTech provides data anonymization and distributed management technology using its "ZENMU Engine" secret sharing engine, while Technica AI is responsible for the design and implementation of RAG and the AI utilization platform. Aegis Application undertakes the organization of customer issues, implementation support, and business promotion for secure AI utilization.

Furthermore, a patent application (Patent Application 2026-39556) for this technology was completed on March 12, 2026.

This solution aims to be provided as a foundational technology for securely utilizing generative AI for customers requiring extremely high security standards, such as those in defense, government agencies, finance, medical care, and manufacturing.

Background: Challenges in data management with conventional RAG

The adoption of RAG (Retrieval-Augmented Generation), which searches and refers to documents to generate answers, is progressing in the utilization of generative AI for internal business operations. However, conventional RAG system configurations have raised concerns about the following risks, which have become barriers to AI utilization.

Data leakage risk

Important information assets accumulated within the company are stored in a search database (vector DB), leading to a high risk of theft due to cyberattacks.

Key risk

Conventional encryption technology requires strict management of encryption keys, and the leakage of the key itself directly leads to the leakage of all data.

Internal fraud risk

Individuals with administrative privileges, such as system administrators, can view the stored data.

"Three Data Protection Functions" realized by this patent-pending technology

This technology integrates ZenmuTech's secret sharing engine "ZENMU Engine" with Technica AI's RAG technology. Secret sharing technology is a method that, instead of "encrypting" data, "transforms it into meaningless distributed fragments (anonymization) and stores them dispersedly in physically different locations."

"Anonymization" and distributed storage to storage

Original data is divided into multiple "shares" (distributed fragments) by secret sharing technology. Each share is stored on physically separate servers, so even if data is stolen from a single server, the original information cannot be restored from a single share.

"On-memory restoration" only during processing and immediate deletion

Distributed fragments of data necessary for AI to generate answers are restored in memory (temporary area) only during processing, and immediately deleted from memory after processing is complete. Therefore, no traces of information are left in the system.

Protection of usage history (prompts and answers)

AI queries, generated answers, and their underlying data are also consistently secret-shared and stored. This reduces the risk of confidential information leaking from past usage history.

Figure 1. Conceptual diagram of Secret Sharing x RAG technology (patent pending)

Superiority of this technology: Establishment of Knowledge Security

This technology realizes "Knowledge Security" for utilizing AI while protecting the important intellectual property (knowledge) of companies and organizations.

Liberation from key management

Physical dispersion overcomes the weakness of conventional encryption, which is the risk of key theft.

Improved tamper resistance

If even a part of the dispersed shares is tampered with, the data cannot be restored, thus providing a self-defense function against data poisoning attacks.

Prevention of viewing by administrators

Even with administrator privileges, the contents cannot be viewed with only a part of the dispersed data, thereby preventing internal fraud.

Future Outlook

From April 2026, the National Institute of Advanced Industrial Science and Technology (AIST) has begun evaluating the safety of this system. Going forward, we will proceed with phased development towards social implementation through patenting procedures, prototype release, and demonstration experiments with early adopter users.

________________________________________

[Patent Application Information]

•Application Date: March 12, 2026

•Application Number: Patent Application 2026-39556

•Title of Invention: Information processing server, information processing system, and retrieval-augmented generation method

Company Overview

Aegis Application Inc.

Aegis Application is a specialist that provides comprehensive consultation and support for customers considering AI adoption. Starting from understanding customer challenges and consulting, we propose, verify, and provide various supports for practical implementation optimized for each customer's environment. For foreign companies.