Elith and Macnica Conduct Japan's First Edge AI Proof of Concept to Detect Special Fraud Calls

Elith and Macnica Conduct Japan's First Edge AI Proof of Concept to Detect Special Fraud Calls

— Confirm Over 90% Detection Accuracy in Real-Time Inference Powered by Broadcom NPU —

Elith Inc. (Headquarters: Bunkyo-ku, Tokyo; CEO: Motoki Inoue; hereinafter "Elith") and Macnica, Inc. (Headquarters: Yokohama, Kanagawa; President: Kazumasa Hara; hereinafter "Macnica") today announced that they have conducted Japan's first*1 Proof of Concept (PoC) using Edge AI to detect special fraud phone calls.

Utilizing a PON MAC SoC*3 evaluation board equipped with Broadcom's NPU (Neural Processing Unit)*2, the two companies established a real-time inference evaluation environment for this PoC and confirmed a detection accuracy of over 90%. This achievement is expected to contribute significantly to the prevention of special fraud damages in landline telephone environments.

Background and Social Issues

According to the National Police Agency, special fraud cases in 2025 reached a record high, with 27,832 reported cases and total damages amounting to 142.31 billion yen*4. Phone calls account for approximately 80% of initial contact methods in special fraud, with a significant portion of damages among individuals aged 60 and older originating from landline phones*5.

Furthermore, in July 2026, the Ministry of Internal Affairs and Communications (MIC) urged telecommunications carriers to further strengthen measures to prevent damages from special fraud. This request included initiatives to raise awareness and promote the adoption of fraud detection and warning services powered by AI for landline telephone users*6.

Conventional nuisance call prevention measures have primarily relied on telephone number identification. However, as fraud methods become increasingly sophisticated, there is a growing demand for measures that can detect early signs of fraud from the content and flow of conversation after a call is answered. On the other hand, analyzing audio in the cloud poses privacy concerns regarding the external transmission of voice data, as well as operational expenses such as server and cloud usage fees.

To address these challenges, Elith and Macnica focused on Edge AI*7—which performs AI inference directly on the device—and conducted this PoC. Elith, with its strengths in Edge AI implementation such as AI model lightweighting and device-level optimization, partnered with Macnica, which possesses deep expertise in semiconductors and telecommunications infrastructure, to jointly drive development leveraging their respective strengths.

PoC Overview and Evaluation Results

This PoC was realized through joint efforts by Elith and Macnica, spanning from AI model development and performance evaluation to edge implementation verification, utilizing Broadcom’s NPU-equipped PON MAC SoC. By combining both companies' AI technology and expertise in telecommunications infrastructure, verification is underway toward the practical application of real-time fraud call detection technology.

Specifically, call audio input via devices such as USB speakerphones is analyzed in real time on the device side within an evaluation environment modeled after home gateways*8, determining signs of special fraud from the conversation content and context (see Figure 1).

Figure 1: Architecture Diagram of Edge AI Fraud Detection PoC

Using publicly available voice data, this PoC evaluated multiple special fraud scenarios—including "Ore Ore" (impersonation) fraud, government official impersonation, and call-center-style fraud—as well as general conversation scenarios. As a result, a detection accuracy of over 90% was confirmed within the evaluation environment constructed for this PoC. This demonstrates the potential of NPU-powered Edge AI to determine fraud risks in real time during phone calls directly on telecommunications equipment.

Note: These results were obtained in a PoC evaluation environment; actual performance may vary depending on real-world usage environments, call quality, conversation content, and other conditions.

Key Features of This Solution

1. Real-Time Fraud Detection Powered by NPU / Edge AIBy performing Edge AI inference processing on terminal-side equipment such as home gateways connected to landlines—rather than in the cloud—call content is analyzed in real time to detect signs of fraudulent phone calls. Utilizing NPU-based processing enables highly efficient AI inference.
2. Privacy Protection for Voice DataWhile cloud-based AI requires sending voice data to external servers, this solution completes all AI processing locally within the device. This contributes to protecting user privacy and reducing the risk of data leaks.
3. Low-Cost Operation with Reduced Cloud ExpensesBy executing AI processing on the edge, server fees and cloud operational costs are significantly reduced compared to cloud-based audio analysis architectures. This creates a cost-effective setup that is easy for telecommunications carriers and service providers to adopt.
4. PoC Designed for Landline EnvironmentsAssuming a setup combining home gateways and landline phones, the PoC verified the feasibility of fraud detection technology that analyzes conversation context in real time using Edge AI. It complements traditional phone-number-based countermeasures and is expected to be applied as a new solution that captures fraud indicators directly from conversation content.

Future Outlook

Based on the insights gained from this PoC, Elith and Macnica will expand evaluation scenarios, improve AI model performance, and advance verification under real-world conditions through collaboration with telecommunications carriers and communication equipment manufacturers.

Additionally, the two companies will evaluate implementation feasibility on telecommunications devices such as home gateways and examine operational challenges, aiming to contribute to a safer and more secure communication environment through special fraud prevention solutions.

Contact Information for Product Inquiries

Macnica, Inc.

Finesse Company — Ohara / Sakai

TEL: +81-45-470-9831

E-mail: bcm_wifi_support@macnica.co.jp

Elith Inc.

Fubuki Sawa

TEL: +81-90-6602-9320

E-mail: fubuki.sawa@elith.ai

About Macnica, Inc.

Macnica is a service and solution company providing a full range of cutting-edge technologies, with semiconductors and cybersecurity at its core. Operating across 100 locations in 33 countries and regions worldwide, Macnica leverages its technical expertise and global network built over more than 50 years to discover, propose, and implement advanced technologies such as AI, IoT, and autonomous driving.

Website: https://www.macnica.co.jp/ (https://www.macnica.co.jp/)

About Elith Inc.

Elith Inc. is a tech company engaged in advanced research and the development of enterprise AI products built on a foundation of AI safety and security. Elith provides products and services that support AI quality, safety, security, and governance in mission-critical operations and decision-making sectors, including manufacturing, finance, healthcare, and social infrastructure. Through practical implementation—building, evaluating, protecting, and continuously improving AI—Elith helps companies build robust foundations to deploy AI in production with confidence.

Business Outline: AI research, development, design, planning, education, sales, maintenance, and consulting services.

Company Profile: https://elith.ai (https://elith.ai)