The Korea Data Agency (K-DATA), led by President Yang Jae-soo, announced that it will strengthen its data quality assurance framework for datasets generated under the 2026 Data Voucher Program to improve their reliability and usability.
The enhanced quality management measures are designed primarily to establish a foundation for securing high-quality datasets under the newly introduced Public Access & Utilization category. The initiative is intended to ensure that data generated through the program can be used not only for enterprise AI service development but also for training a wide range of AI models.
The Data Voucher Program, jointly administered by the Ministry of Science and ICT and K-DATA, supports data-driven AI service development and commercialization for small and medium-sized enterprises (SMEs), youth-led businesses, prospective entrepreneurs, and early-stage startups.
In 2026, the program will support 120 projects across two categories—General and Public Access & Utilization—with a total budget of KRW 7.2 billion.
General category: 60 projects, approximately KRW 45 million per project
Public Access & Utilization category: 60 projects, approximately KRW 75 million per project
The newly established Public Access & Utilization category is designed to make datasets created through the program publicly available, enabling other companies and research institutions to reuse them for AI development and service enhancement.
To support the introduction of this category, K-DATA will implement a customized quality inspection system tailored to different data types, together with a multi-layered quality management framework aimed at securing datasets suitable for AI training and data reuse.
The quality assurance process will cover structured, semi-structured, and unstructured datasets produced through the Data Voucher Program.
Structured data: Data organized in tabular formats, including master data, transaction data, and aggregated data.
Semi-structured data: File-based data containing structural information, such as HTML, XML, and GIS data.
Unstructured data: Data without predefined structures, including videos, images, audio, and documents.
For structured and semi-structured datasets, K-DATA will utilize automated quality inspection tools aligned with national public data quality management standards to examine table structures, data standards, metadata, and other quality indicators. The objective is to improve data standardization and interoperability while ensuring that datasets can be readily utilized across diverse systems.
For unstructured data, the agency will expand its quality assessment beyond the previous image-focused approach by applying dedicated quality criteria to images, videos, audio files, and documents. In particular, automated quality diagnostics will be extended to video and audio data, which had previously relied largely on manual visual inspection, improving both the accuracy and efficiency of quality evaluation.
K-DATA will also establish a multi-layered quality management framework that integrates supplier self-assessment, automated quality diagnostics, and expert quality reviews. Through this step-by-step verification process, the agency aims to secure high-quality datasets suitable for AI training and data reuse while systematically managing data reliability and usability.
The strengthened quality management framework is expected to expand the practical use of datasets created through the Data Voucher Program and further reinforce the availability of high-quality data essential for AI service development.
"AI competitiveness ultimately begins with high-quality data," said Yang Jae-soo, President of K-DATA. "We will continue to advance our data quality management framework so that datasets produced through the Data Voucher Program can be trusted and utilized by everyone, thereby actively supporting enterprise AI transformation and data-driven innovation."
