Ministry of Labor Deploys AI for Industrial Accident Prevention: "Prediction Performance Improves by 52%"

The government is set to actively promote an "Artificial Intelligence Transformation (AX)" administrative model that leverages artificial intelligence (AI) to predict industrial accidents and enhance the efficiency of labor administration.

On March 13, the Ministry of Employment and Labor held the "Our Ministry of Labor AX Seminar," unveiling cases in which public officials directly used AI to achieve administrative innovation. The Department of Labor Administration Artificial Intelligence Innovation developed an initial AI model to predict the risk of industrial accidents, utilizing AI coding agents such as "Claude Code" and "Codex." This model is trained on data such as the history of industrial accidents and supervision records from approximately 3 million workplaces nationwide, enabling the precise identification of about 19,000 high-risk workplaces, which account for the top 0.6% with the highest probability of accidents.

The Ministry expects that by focusing limited industrial accident prevention personnel and budgets on the selected workplaces, the effect of reducing industrial accidents can be maximized. In particular, to ensure the accountability and transparency of administration, the model was designed as "explainable AI," making it possible to logically present the reason why a particular workplace was chosen.

Workers are performing rebar work in Changneung District, Donaedong, Deogyang-gu, Goyang City, Gyeonggi Province. Photo by Yonhap News Agency

The performance evaluation results also showed significant improvement over conventional methods. Compared to the existing system, where people assign scores based on industrial accident history and other factors to select workplaces subject to management, the prediction performance improved by 52%. Analysis based on data available as of December 2024 showed that, among the 3 million workplaces, 19,000 were selected. In workplaces selected by humans, the number of lost workdays due to occupational injuries in 2025 was 1.93 million days, whereas in those selected by AI, it was 2.94 million days. This indicates that AI was more accurate in identifying workplaces with a high risk of industrial accidents.

A work support tool developed directly by field public officials was also released. "Sound Writer," developed by Kang Min, a labor inspector at the Busan Regional Employment and Labor Office, is a tool that instantly converts call recording files obtained during investigations into text using AI. It is expected to help expedite the identification of the facts of a case and facilitate the protection of workers' rights by enabling rapid analysis of vast amounts of recorded material.

Han Isong, an official at the Seoul Western Employment Center, developed the "AI Cover Letter Generator," which assists in drafting tailored cover letters based on job seekers' backgrounds and competencies and is currently being used in the field. This tool, which helps job seekers write persuasive cover letters when applying to multiple companies, is reportedly receiving positive feedback among vocational counselors.

The Ministry plans to thoroughly review the practical applicability of the initial industrial accident prediction AI model and continuously enhance its predictive performance. In addition, it intends to launch full-scale development of "wage arrears prediction AI" to proactively manage workplaces at risk of wage non-payment. The Ministry will also expand support and training for public officials in using AI coding agents, so that outstanding cases developed in the field can be disseminated nationwide.

Minister of Employment and Labor Kim Younghoon stated, "We will accelerate the AI transformation so that every employee can skillfully use AI to resolve wage arrears cases as swiftly as possible, and offer pinpoint consulting and budget support to workplaces vulnerable to industrial accidents."

This content was produced with the assistance of AI translation services.