Is AI to blame for youth unemployment? Budget office says it's too soon to tell

Blaming generative AI for the recent slump in youth employment is a premature conclusion, according to a new analysis. While hiring and employment of young white-collar workers have been relatively weak, the data show no consistent pattern in which higher AI exposure leads to steeper declines in youth employment.

Rather than treating AI exposure alone as a reliable indicator of employment risk, the analysis recommends continuously tracking actual AI use alongside changes in hiring, employment and income — and responding in stages only as risks are confirmed.

The National Assembly Budget Office published a report titled "The Impact of Generative AI on the Labor Market," the office said Sunday.

The report measured AI exposure by occupation based on actual workplace use, and examined whether differences in employment, hiring and wages emerged across exposure levels — all amid the rapid spread of generative AI in the workplace. According to earlier research, 51.8 percent of domestic workers used generative AI on the job in the first half of 2025, a higher rate than in the United States, where the figure stood at 45.9 percent.

The budget office separately surveyed 1,512 workers across 57 predominantly white-collar occupations. Of those surveyed, 76.1 percent said they used generative AI at work, and AI users reported an average 21.2 percent increase in their workload capacity after adopting the technology.

Using the survey results to calculate average productivity gains by occupation, the office classified IT specialists, researchers, and finance and legal professionals as high-exposure occupations.

Comparing those exposure levels against actual employment trends revealed differences by age group. Overall employment among workers aged 20 to 59, measured by employment insurance enrollment, followed a gradual upward trend after ChatGPT launched in November 2022. Employment among workers in their 20s, however, declined across white-collar occupations — excluding those with no AI exposure. Within the white-collar category, no consistent relationship emerged between the degree of AI exposure and the pace of employment decline.

Youth hiring also showed a broad downward trend. The number of new employment insurance enrollees slowed its growth from the second half of 2021, turned negative in the second half of 2022 and continued to fall. After ChatGPT's launch, hiring of young workers in high-exposure occupations appeared relatively weaker, but an overall decline in youth hiring was observed at the same time.

The budget office cautioned, however, that these trends alone are not sufficient to attribute the employment and hiring declines to generative AI. The slowdown in sectors such as information and communications technology and professional and scientific services — which employ a large share of high-exposure workers — could be mistaken for an AI-driven effect, the office said.

After accounting for sector-level economic conditions, the office analyzed changes in employment, hiring and wages by AI exposure level before and after ChatGPT's launch. For both the 20–59 age group and workers in their 20s, no statistically significant effect of generative AI on employment in high-exposure occupations was found. The same held when comparing high-exposure jobs against low-exposure or unexposed ones. Wages also showed no significant impact.

Results for hiring varied depending on the comparison group. High-exposure occupations showed a statistically significant positive effect on hiring when compared with low-exposure jobs, but a non-significant negative effect when compared with unexposed ones. The office concluded that the direction of generative AI's impact on hiring remains uncertain.

Taken together, the findings led the budget office to conclude that the recent relative weakness in white-collar youth employment and hiring is difficult to attribute causally to generative AI.

Not only did higher AI exposure within the white-collar sector fail to produce steeper youth employment declines, but differences in employment and hiring trends across exposure levels could also be explained by industry-level factors — such as the post-pandemic slowdown in demand for non-face-to-face services.

The office cautioned, however, that these findings alone are not sufficient to conclude that generative AI has had no effect on the labor market. The impact of AI's spread may take time to materialize, and some changes may not yet be captured by available statistics.

The office recommended treating AI exposure as a screening indicator for monitoring rather than a risk verdict, and tracking actual AI use alongside changes in employment, hiring and income. Under the proposed framework, workers would receive general AI skills training if no abnormal changes appear; if hiring or income declines recur, deeper investigations and employment impact assessments would follow; and if negative trends persist, the response would expand to include job transition support, career retraining and income assistance.

The office also proposed linking corporate AI transformation support to on-the-job retraining programs and strengthening assistance for small and medium-sized enterprises. For young workers, it called for reinforcing early career pathways connecting training to work experience and employment. For non-standard workers, it said monitoring should extend beyond layoffs and unemployment to cover reductions in work volume and income as well.