SEOUL — In a strategic move to solidify the nation’s standing in the global artificial intelligence race, the South Korean government has announced its first-ever comprehensive census of AI training data held by the public sector.
The Ministry of Science and ICT (MSIT), in collaboration with the National Information Society Agency (NIA), launched the "AI Training Data Census" project on April 10, 2026. The initiative is designed to systematically identify, evaluate, and integrate high-quality datasets scattered across various government ministries and public agencies into a centralized accessible platform.
Bridging the "Data Divide"
Despite the wealth of information collected by public institutions, a significant portion of AI-ready data currently remains "siloed." Because these datasets are managed independently by different organizations, private AI developers and startups have long faced challenges in identifying what data exists and how to utilize it for machine learning.
"The core of AI performance and quality lies in the availability of abundant, high-quality data," said Kim Kyung-man, Director General of the Artificial Intelligence Policy Bureau at MSIT. "Through this census, we will systematically discover the AI data assets held by the public sector and build a foundation where they can be utilized conveniently by the private sector."
The 100 Dataset Initiative
The project is rooted in Article 15 of the AI Basic Act, which mandates the establishment of policies to promote AI training data. Key highlights of the roadmap include:
Comprehensive Inventory: The census will cover not only existing AI-ready data but also raw data that has high potential for future processing and labeling.
Selection of High-Value Assets: The Ministry plans to identify 100 priority datasets that show the highest potential for industrial and social utility.
Technical and Financial Support: A budget of 6 billion KRW has been allocated to the "Integrated AI Training Data Provision System." This funding will support the cleaning, standardization, and de-identification (anonymization) of the selected 100 datasets to ensure they meet strict privacy and quality standards.
Infrastructure Expansion: These datasets will be integrated into 'AI Hub,' a national platform that already hosts 903 types of data across 14 categories.
Ensuring Security and Accessibility
Recognizing that some public data may contain sensitive information that cannot be shared openly online, the Ministry announced a hybrid distribution model. While most data will be available for download via the revamped AI Hub, highly sensitive or restricted datasets will be accessible through 'Data Safe Zones'—secure physical or virtual environments where researchers can train models without the risk of data leakage.
The census will evaluate data based on standardized metrics, including data structure, purpose of construction, and the scope of legal provision. To ensure the selection process reflects real-world market needs, the Ministry will conduct in-depth expert interviews and private-sector demand surveys.
A Foundation for Sovereignty
Experts view this census as a critical step toward "AI Sovereignty." By unlocking public data in sectors such as healthcare, administrative services, and urban planning, South Korea aims to help local tech companies reduce their reliance on foreign datasets and build AI models that are better optimized for the Korean language and cultural context.
As the global competition for high-quality "tokens" (units of data used for AI training) intensifies, South Korea’s proactive approach in treating public data as a national strategic asset is expected to provide a significant tailwind for the domestic AI industry.
[Copyright (c) Global Economic Times. All Rights Reserved.]
