US Patent Acquired for Multilingual Document Analysis Method for Real-time Detection of Threats, Fraud, and Errors in Outgoing Emails

iMatrix Corporation (Headquarters: Kawasaki City, Kanagawa Prefecture; Representative Director: Akira Oka) is pleased to announce the acquisition of a US patent (Patent No.: US12,602,542B2) for its core technology in "matriXinsight," an anomaly detection and monitoring solution for outgoing emails. This technology, a method for extracting and analyzing features from email bodies, allows for the learning of sender's writing characteristics regardless of the email body's language, making it possible to detect emails written in topics or styles that the sender does not usually handle. We believe that matriXinsight, utilizing this technology, can contribute to strengthening the security environment of Japanese companies struggling with a shortage of security personnel, as it enables real-time outgoing email checks without the need for explicit keyword settings, tuning, or AI management.

＊＊＊＊＊＊＊＊＊＊＊＊＊＊

matriXinsight is a real-time outgoing email monitoring service available within "imatrix Security Platform," an email security service provided by iMatrix.

＊＊＊＊＊＊＊＊＊＊＊＊＊＊

Figure 1: Overview of matriXinsight

■ Background

While chat tools are becoming mainstream for internal business communication, email remains the primary method for communication between different companies. Challenges in inter-company communication include information leakage and compliance violations. Although functions exist to prevent misdelivery, such as send-hold and recipient confirmation tools, and third-party check tools requiring manager approval, implementing these for all emails is impractical due to high time and effort costs. Therefore, in most cases, responses are limited by conditions, but setting minimum rules without omissions often requires significant effort or is impossible. Some cases adopt a method of sampling some outgoing emails and checking them manually, but this is a post-check and only a portion of emails can be checked.

■ Real-time Analysis of Outgoing Emails by matriXinsight

matriXinsight is a solution designed for organizations and departments facing the aforementioned challenges, to carefully select emails sent externally that human attention should particularly focus on for review. iMatrix's proprietary AI automatically learns outgoing emails for each individual sender, defining "emails usually sent by that sender." It can then issue alerts, record, or request manager approval for emails that deviate from this definition.

Generally, misdeliveries, fraud, or account takeovers often involve clearly unusual behavior, and matriXinsight scores these "unusual characteristics." The greater the difference from usual emails, the higher the score. Scoring is performed in real-time during the email sending process, allowing anomalies to be detected before the email is sent externally.

Administrators and compliance check personnel can focus more on their core checking duties by observing emails with high scores from matriXinsight, without having to visually check all outgoing emails. This also allows for pre-emptive, on-the-spot checks instead of post-checks, increasing the possibility of preventing information leakage and fraud.

matriXinsight scores are assigned to the header, body, and attachments, indicating which parts should be prioritized for checking. The newly acquired patent is utilized in the scoring of the email body, enabling analysis and feature extraction/scoring using the same method even when the email body is written in languages other than Japanese or in mixed languages. It can be used without special tuning even in departments or group companies that frequently interact with overseas entities.

■ Fully Independent Learning Model Without External Models

matriXinsight's AI creates a dedicated learning model for each sender using only the outgoing emails of that user, without using public models or linked functions from external organizations. It does not use other users' data for learning or inference, nor vice versa. Since learning data is not shared externally, there are no concerns about data leakage due to AI use or the mixing of unauthorized data into the learning model.

Learning is performed automatically, and features are regularly updated from accumulated new outgoing emails. Therefore, it can be operated even if the administrator is not an AI expert.

＊＊＊＊＊＊＊＊＊＊＊＊＊＊

Reference price range: Approximately ¥210/account/month
Keywords: