【Interview #16】KeiganAI - Safeguarding Information Integrity in the AI Era. A PM’s Vision for Social Implementation

We interviewed Mr. Suzuki, who serves as the Head of the R&D Division at NABLAS, overseeing a wide range of development projects across the organization. In addition to this broad management role, he also leads development as the PM of the KeiganAI project, promoting its social implementation through partnerships and new business initiatives.

What is the KeiganAI Project?

In today’s era of generative AI, massive amounts of information are being created and circulated across social media and the internet. Among them, malicious content created with deepfakes and information containing hallucinations can negatively impact individuals, companies, governments, and even entire nations.

To address this social challenge, the KeiganAI project develops technologies for fake detection and fact-checking. At NABLAS, we are advancing these technologies as safeguards for the safety and reliability of information, and we are also working on fact-checking AI agents capable of conducting such tasks autonomously.

Could you tell us about the work you’ve recently been focusing on within the KeiganAI project?

As someone leading the project while keeping an eye on the entire division, my role is not only on the development side but also in partner selection and business launches that connect development to real-world implementation.

On the development side, we are working to improve the accuracy of our fake detection models, enhance the explainability of results, and refine the UI/UX to meet the needs of various industries. Beyond distinguishing between “real” and “fake,” we are also developing fact-checking technologies that can determine whether information is true or false.

Currently, we are advancing the development of autonomous AI agents that evaluate not only the content of information itself but also the source and the way society perceives it. This effort is part of the GENIAC project, supported by NEDO under Japan’s Ministry of Economy, Trade and Industry.

Since KeiganAI directly addresses the rising societal challenge of information reliability, we are also focusing on social implementation. Recently, in collaboration with a major telecommunications company and under a pilot project by Japan’s Ministry of Internal Affairs and Communications, we have been developing fake voice detection technology for telephone environments, as well as technologies to prove when information is accurate.

What are the challenges in advancing the project?

One of the biggest challenges in KeiganAI development is the lack of well-organized training data. In AI development, data quality is paramount, but in the field of fake detection there are no high-quality, open evaluation datasets. Additionally, depending on the recording device, default filters or processing can be applied, causing judgments of “real” versus “fake” to vary by use case. This makes evaluation complex.

Moreover, since we are dealing with data and information generated in real life, we need to establish evaluation criteria from scratch while technology itself is evolving rapidly. This is a significant challenge.

Another challenge is that although fake detection is often grouped as one concept, implementation methods and UI/UX requirements differ by industry and use case. Adapting the technology into models that match actual needs through on-the-ground data and feedback is a difficult but necessary task.

Finally, because this is a “detection” technology against “generation,” keeping up with the pace of generative AI advancements is also a challenge. Generative AI is evolving extremely quickly, so maintaining and improving detection capabilities to match these advancements is essential.

How are you addressing these challenges?

The technology must function effectively in real-world environments. That’s why we work closely with industry professionals and institutional partners, engaging in repeated discussions and feedback loops. Their input helps us understand what kind of detection or fact-checking is needed and for which data, and based on that, we develop tailored datasets and detection models.

As for catching up with generative technologies, we make it a priority to stay updated with the latest developments across modalities. As AI professionals, we monitor cutting-edge generation technologies and develop our own original models, while maintaining a fast implementation cycle to keep our detection capabilities up to date.

 What do you find most rewarding about this project?

As AI adoption grows worldwide, I believe that the fake problem is a challenge everyone will eventually face. Delivering technology to solve this issue to society is a tremendous source of fulfillment.

The fake problem, in the context of “information,” has significant influence and can even impact balances at the national level. While this is a large-scale issue, I believe our detection technology helps safeguard the integrity of information. In that sense, I feel proud to be working on technology with a high degree of social contribution.

This also ties into NABLAS’s vision of “Be a good ancestor.” Implementing KeiganAI, a technology for detecting fakes, as foundational infrastructure for a better future society is, I believe, truly meaningful.

What are your future goals and challenges?

My goal is not only to determine whether information is fake or not, but also to develop technology that can assess whether the information itself is genuinely trustworthy. I want to create technology that enables everyone to explore and use information fairly.

Furthermore, this is not something meant to stay within Japan—it should transcend languages and borders to be used worldwide. If we can build robust technology domestically, I am confident it will have global relevance. Ultimately, I want KeiganAI to become a service that is widely used across the globe.

NABLAS is currently recruiting new members as we expand our business!