Through the K-AI Alliance, SK Telecom is joining forces with leading Korean AI startups to strengthen the domestic AI ecosystem and global competitiveness. More than 30 innovative companies are currently participating in the alliance, collaborating across initiatives including the Ministry of Science and ICT’s Proprietary AI Foundation Model Development Project.
As part of SKT’s core team in the Proprietary AI Foundation Model initiative, Datumo(known as Selectstar in Korea) is at the forefront of generative AI reliability verification through its solution “Datumo Eval”. Founded in 2018, the company has built Korea’s first all-in-one reliability verification solution and has presented at major global AI conferences. In the consortium, Datumo is responsible for overseeing the data domain, including building red-teaming datasets for model stability verification and developing high-quality training data.
SKT Newsroom met with David Kim, CEO of Datumo, to find out more about the outcomes of participating in the K-AI Alliance and the role the company is playing in the SKT consortium.
* This article contains the views of another company and may not represent SK Telecom’s official position.
“Datumo Eval” – Korea’s First AI Reliability Verification Solution
Datumo CEO Seyeob David Kim
Datumo is a company specializing in AI dataset development and reliability verification, founded in 2018 by a startup team out of KAIST. With the mission of “changing the world with AI,” our company has grown by focusing on data, the core competitiveness of AI. In just seven years, Datumo has secured more than 320 customers, supplied over 250 million high-quality datasets, and established itself as a leading AI company.
Recently, we were selected as one of “Forbes Asia 100 to Watch”, confirming our potential on the global stage. Datumo is also participating in the Ministry of Science and ICT’s Proprietary AI Foundation Model project as a core team of SKT’s consortium. As a data-specialized company, Datumo is taking on the dual responsibilities of “creating datasets” and “verifying reliability” throughout the AI model lifecycle, based on our accumulated expertise.
Q. Why is AI service reliability verification important?
With the growing integration of AI into daily life and diverse industries, ensuring that AI outputs are accurate, fair, and safe has become a critical challenge. Companies are indeed enhancing productivity and efficiency through AI, but they are also facing risks such as hallucination, where AI produces false information, or the generation of unintended biased or harmful content.
Such risks can lead to reputational damage and legal issues for companies. In sectors like finance, healthcare, or public services where small errors can cause major harm, these risks become even more critical. Reliability verification is the key process that detects and mitigates such risks preemptively, making the safe operation of AI services possible.
Traditional AI evaluation methods often relied on the subjective judgment of experts or unstandardized criteria. Datumo has addressed this by establishing a framework that quantitatively verifies whether AI functions accurately, avoids harmful outputs, and aligns with policy and societal standards. This framework is the foundation of our AI reliability verification solution, Datumo Eval.
Datumo’s “Datumo Eval” service interface
Q. What are the features of Datumo’s reliability verification solution, Datumo Eval?
Datumo Eval is Korea’s first solution that automatically evaluates the reliability of generative AI. It automatically generates evaluation question sets across various domains and scenarios, then evaluates AI responses against predefined standards. Quantitative results are displayed on a dashboard, and it also offers automated red-teaming functions that deliberately expose vulnerabilities.
Datumo Eval’s greatest strengths are precision and efficiency. Evaluation processes that used to take over a month can now be completed in a single day. The solution applies “agent-based workflowsUsing multiple LLM agents in a distributed process for evaluation” and “human alignment technologiesTraining AI systems to align their outputs with human standards (values, preferences, and safety)
” to significantly enhance reliability. It also allows the setting of customized criteria according to industry regulations or policies, making it applicable across diverse fields. Currently, Datumo Eval is the only service in the global market capable of performing the full cycle, including automated generation of evaluation data.
Spearheading Korean AI Models’ Quality as Data Domain Leader in the SKT Consortium
Q. What are your thoughts on participating as a K-AI Alliance partner?
Datumo joined SK Telecom’s consortium for the Proprietary AI Foundation Model initiative and became a member of the K-AI Alliance. The Alliance is a symbolic initiative that demonstrates the strength of Korea’s AI technology, going beyond a simple corporate network. Especially under SKT’s global leadership, companies with diverse expertise are working together, creating opportunities for synergy through technological convergence and joint planning. As a participating company, we expect not only to contribute to the growth of Korea’s AI ecosystem but also to showcase Datumo’s strengths in data and reliability verification.
Q. What was the background behind joining SKT’s Proprietary AI Foundation Model consortium?
As the global AI market grows rapidly, the need for Korea to build its own AI foundation model has become urgent in order to maintain competitiveness. I believe this project represents new industrial sovereignty in the AI era and will be key to national competitiveness. Datumo joined with the conviction that, as a data-focused company representing Korea, we could contribute with our experience across various industries in dataset creation and reliability verification technologies.
This consortium is particularly meaningful because it is designed not only for technical development but also with global expansion in mind. With SKT leading the initiative, we expect the consortium will achieve a user-friendly model and realize a full-stack AI built on proprietary technology.
Q. What role is Datumo playing within the consortium?
Datumo is spearheading the data domain. Based on experience building over 250 million datasets, we support the entire cycle from training to evaluation.
First, using our “CashMissionDatumo’s data processing platform with 240,000 contributors” platform, we will transform real-world data and accumulated technical know-how into high-quality training datasets, and create datasets required for multimodal model training. Additionally, we aim to provide an advanced framework capable of testing safety and bias using our Datumo Eval solution, the Korean-language benchmark KorNAT, and our experience operating red-teaming challenges for generative AI domestically and internationally. This helps identify potential vulnerabilities in advance and strengthens model stability.
Q. What are your views on SKT’s vision to evolve into a “Global AI Company”?
SKT’s vision goes beyond adopting AI technology – it intends to embed AI throughout operations, services, products, infrastructure, and culture. With its assets in telecommunications, data, infrastructure, and platforms, SKT is uniquely positioned to translate AI technology into real customer experiences and services. This foundation, when linked with Korean AI startups, can expand into a stronger ecosystem, providing powerful motivation for K-AI Alliance members to pursue greater goals.
Datumo CEO Seyeob David Kim
Q. What contribution does Datumo hope to make in building a Korean AI ecosystem?
Our goal is to establish objective and transparent verification systems for AI outputs so that AI becomes recognized as a trustworthy technology. An AI ecosystem is complete only when data, models, and services are organically interconnected. With our expertise in data and model verification, Datumo aims to contribute to building a more reliable AI ecosystem.
The foundation of AI technology is ultimately high-quality data. We secure legitimate data that is difficult to obtain from the web, such as expert texts, and manage the entire lifecycle from design, processing, and delivery to ensure quality. Importantly, we provide datasets that address complex issues such as privacy, copyright, and policy compliance, enabling Korean companies to safely develop and utilize AI.
We intend to take the lead in addressing the reliability issues that are the most critical challenge for AI commercialization. Through Datumo Eval, we will help companies operate AI more accurately and safely, while supporting the early identification and resolution of potential errors and risks.
Q. What are your plans for the future?
Datumo aims to become a global AI infrastructure company leading the world in AI dataset creation and reliability verification. In the long term, we want to be recognized as the “standard for trustworthy AI” and a central player in advancing the quality of Korea’s AI ecosystem.
As part of these efforts, we will strive to be selected in the final stage of the Proprietary AI Foundation Model project. We aim to contribute to SKT’s consortium in realizing a true “national AI” that all Koreans can use, while focusing on strengthening Korea’s AI competitiveness.
We also plan to accelerate our global expansion. By globalizing Datumo Eval as a SaaS, we will enable overseas companies to easily perform reliability verification, and we will expand international dataset collaborations starting with the North American market. At MWC26 next year, we will host a larger global red-teaming challenge than this year to continue raising global awareness of AI reliability verification.
