[2026 National Assembly Audit] AI-Predicted High-Risk Workplace Inspection Rates: 63.8% in Construction, 5.2% in Manufacturing

37.1% Inspection Rate at Construction Sites Worth KRW 12 Billion or More
Park Jeong: "Reflect AI Predictions in Field Inspections and Upgrade the System"

As the government moves to establish an AI-based system to prevent industrial accidents, a significant gap has emerged between workplaces identified by AI as high-risk and those actually selected for on-site inspections.

Reuters-Yonhap News

According to data submitted by the Korea Occupational Safety and Health Agency, Democratic Party of Korea lawmaker Park Jeong, a member of the National Assembly’s Climate, Energy, Environment and Labor Committee, said on October 28 that 15,584 of the 24,418 construction sites identified by AI as high-risk from January to September this year were subject to supervision or inspection, for an inspection rate of 63.8%.

Among construction sites, those with construction costs of KRW 12 billion or more accounted for the largest number of high-risk sites. Only 3,419 of the 9,216 such sites were inspected, leaving the inspection rate at 37.1%. By contrast, sites with construction costs of KRW 2 billion to less than KRW 5 billion had the highest inspection rate, at 88.4%.

The inspection rate was even lower in manufacturing and other industries. Of the 373,567 workplaces identified by AI as high-risk, only 19,402 were actually inspected, for an inspection rate of just 5.2%. The number of high-risk workplaces was about 15 times that in construction, but the number of workplaces inspected was only about 1.2 times as high.

In manufacturing and other industries, workplaces with 5 to fewer than 50 employees made up the largest high-risk group, but their inspection rate was 6.0%. Workplaces with 300 or more employees, which had the fewest high-risk sites, had the highest inspection rate: 749 of 3,394 were inspected, for a rate of 22.1%. The inspection rate for workplaces with fewer than five employees was 1.4%.

The Korea Occupational Safety and Health Agency operates a system that uses AI to learn from data on about 200 indicators and calculate the risk level of each workplace, thereby predicting which workplaces are high-risk.

“Even when AI predicts which workplaces are high-risk, the predictions do not translate into actual on-site inspections,” Park said. “We need to carefully determine whether the predictions are not being properly reflected in the field or whether the prediction system itself needs to be upgraded, so that inspections can focus on workplaces where the actual risk is high.”

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