neoAI implements advanced auditing functions in its generative AI platform “neoAI Chat” – Ikeda Senshu Holdings decides to adopt it.

neoAI Inc. has implemented advanced auditing functions in its generation AI platform “neoAI Chat” to ensure the safe and appropriate use of generation AI, and is pleased to announce that it has been decided to introduce it to Ikeda Senshu Holdings Co., Ltd.

“neoAI Chat Advanced Audit Function” is an optional feature for “neoAI Chat” designed for corporate use, offering guardrail and audit log acquisition functions. It supports internal controls, incident investigations, and audit responses through its guardrail function, which monitors user input and prevents risky transmissions, and its audit log acquisition function, which records system authentication and operations within the system.

As the full-scale adoption of “neoAI Chat” by companies progressed, neoAI recognized the growing need for enhanced governance of its generative AI. In particular, companies with high security standards, such as financial institutions, required control over the risk of users unintentionally entering information that did not comply with policies, and the acquisition of audit trails to periodically verify that the service was being used securely.

To address these challenges, neoAI has begun working with Ikeda Senshu Holdings to build an advanced auditing function that can meet the company’s standards.

neoAI conducted a Proof of Concept (PoC) with Ikeda Senshu Holdings, discussing the ideal audit approach while refining the detection targets and operational rules for guardrail functions tailored to the context of banking operations, and improving the accuracy of judgments. The PoC primarily focused on the following three aspects:

The guardrail function automatically detects and controls the input and inappropriate use of information that does not comply with the organization’s policies.

To accommodate the different security requirements and operational policies of each financial institution, neoAI is designed to allow for the optimization of guardrail functions for each organization, enabling operation in accordance with their own policies.

To fulfill our accountability to the audit department and management, we have started recording logs of when the guardrail function is activated, so that we can understand usage patterns and use them in audit responses.

In addition, the PoC verified the detection accuracy from the following perspectives:

Based on the results of the proof-of-concept (PoC), we achieved a detection rate of 97.6% and a false positive rate of 0.7%. We have decided to implement this function with a view to actual operation at Ikeda Senshu Holdings, and are proceeding with preparations for the start of operation.