Elith Inc. announces that its research theme, "Development of ‘FinGuard,’ an AI Guardrail Model Specialized for Financial Operations," proposed jointly with Chubu University, has been selected in the 4th call for proposals of "GENIAC (Generative AI Accelerator Challenge)," a project run by the Ministry of Economy, Trade and Industry (METI) and the National Research and Development Agency New Energy and Industrial Technology Development Organization (hereinafter "NEDO") to strengthen Japan’s domestic generative AI development capabilities.
GENIAC is a national project that supports the development of competitive generative AI foundation models within Japan. It is being promoted with the aim of strengthening Japan-originated generative AI technology and accelerating its industrial application.
The "safety" required of financial AI
In recent years, while the use of generative AI has been rapidly advancing in the financial industry,
- Leakage of personal information (PII)
- Compliance violations
- Generation of misinformation (hallucination)
- Attacks such as prompt injection
responding to these and other risks specific to the financial industry has become a challenge.
In this project, we will work jointly with Chubu University to develop "FinGuard," a Japanese-language AI guardrail model specialized for the financial sector.
Features of "FinGuard"
“FinGuard,” which we are developing, is an AI guardrail model specialized for financial operations that monitors the input and output of generative AI in real time, detecting and blocking inappropriate responses and leaks of confidential information.
Its main features are as follows.
- A Japanese-language AI guardrail model specialized for the financial sector
- Detection of personal information (PII) leaks
- Detection of compliance violations
- Detection of financial hallucinations
- Prompt injection countermeasures
- Cross-cutting connectivity to LLMs and AI agents
Rather than being dedicated to a specific service, it is designed for use as a “general-purpose safety layer” that can connect across the various LLMs and AI agents used by financial institutions. This project will also work on building and publishing "FinGuard-Bench," a safety benchmark specialized for the financial domain.
Future plans
Through this project, we will also carry out demonstrations with financial education institutions, financial DX companies, robo-advisor operators, and others, aiming to realize the safe use of generative AI in the financial sector.
Elith will continue to advance research and development in the AI safety domain, which is essential for the real-world deployment of generative AI, and work to build the foundational technology that supports safe and practical AI use.
Page announcing the selection results:https://www.nedo.go.jp/koubo/CD3_100421.html
*About GENIAC
GENIAC (Generative AI Accelerator Challenge) is a project run by METI and NEDO to strengthen Japan’s domestic generative AI development capabilities.
It mainly provides computing resources for the development of foundation models, the core technology of generative AI, and supports demonstration studies for the utilization of data and AI.
This call for proposals aims mainly to accelerate the development of competitive generative AI foundation models through the provision of computing resources, and marks the fourth phase since the project began in February 2024.
URL:https://www.meti.go.jp/policy/mono_info_service/geniac/index.html