Although the government is expanding AI (artificial intelligence) development in the public sector, critics point out that it fails to fully consider the burden on on-the-ground ministries responsible for operating actual services after development. AI requires updates whenever laws, regulations, or administrative data change, and generative AI continuously incurs costs based on usage volume. This means that simply investing in development funds to build a service is not enough.
According to the government on the 8th, generative AI has strong characteristics of a "live service" requiring ongoing management even after development, unlike general information systems. When laws or administrative guidelines change, related data must be reflected again, and token costs are incurred depending on usage volume. To maintain accuracy and performance, model improvements must also continue.
The government is rapidly expanding AI utilization. According to the 2027 budget plan prepared by the Ministry of Employment and Labor, 1.7 billion won for an AI employment center and 3.8 billion won for an industrial safety AI assistant have been included. The industrial safety AI assistant is a system that comprehensively analyzes industrial safety data, risk information, laws, and guidelines to support all stages of preventive projects, including inspections, technical support, and post-management.
Existing AI labor law consultation and self-inspection programs for small businesses were integrated into the 2027 budget plan, increasing related funding from 2.8 billion won this year to 4.4 billion won next year. Additionally, 800 million won for an AI disaster investigation assistant and 600 million won for building an AI-based telephone consultation STT system were newly included.
As the number of AI services increases, so do management costs after development. AI labor law consultations require updating related information whenever laws and guidelines change, and industrial safety AI must continuously update laws, guidelines, and risk information. From the Ministry of Labor's perspective, introducing more AI services means managing an increasing number of systems.
In cases where AI is utilized within closed networks, operational burdens can become even greater. For example, when handling administrative data containing personal information like the Ministry of Labor's closed-type language model "Dawoori GPT," it becomes difficult to directly use publicly available external AI models. Specialized infrastructure such as high-performance graphics processing units (GPUs) is required, and continuous costs are incurred for leasing or maintaining them.
This issue is not limited to the Ministry of Labor. As the government expands AI utilization in the public sector, AI development and introduction projects across various ministries are increasing, but how to secure budgets for stable service operation for several years after development remains a challenge.
In fact, multiple ministries are developing and piloting AI solutions under the government's public AI transformation (AX) project. The 2026 Public AX Project has identified and launched 20 AI convergence tasks, including an industrial injury compensation AI platform and a labor law compliance support solution based on labor supervision AI.
The problem arises after the project ends. If operating costs cannot be stably secured after on-site implementation, even high-performance AI may become Chae (President) underutilized. In particular, AI is not a system developed and completed within one or two years; it is a service that requires updating laws, guidelines, data, and managing models even in the Nth year. There is growing concern that not only the initial development budget and project period but also ongoing operation and enhancement costs must be continuously guaranteed.
Industry experts argue that the root of the problem lies in the government's AI project structure, which separates the development phase from the on-site operation phase. Creating AI by investing development funds and ensuring its continued use on the ground are separate issues. Critics note that if operating funds dry up, development costs may effectively become sunk costs.
As the government promotes the AI Daejeon initiative and public sector AI services expand, voices are growing for establishing a lifecycle budget structure that supports not only development and pilot testing but also post-implementation operation and enhancement on the ground.
in ECONOMY
