In July 2026, South Korea's unemployment rate for ages 15-29 was 6.8%, up 1.3 percentage points from the same month a year earlier. Measured by the size of the increase, it is the largest in five years and six months, since the 1.8 percentage point rise of January 2021. In the same month, youth employment fell by 191,000, a decline for the 45th consecutive month, and the employment rate came down to 44.2%, falling for two years and three months. Yet total employment among those aged 15 and over rose by 108,000.
The two statistics move in opposite directions. It is a picture in which overall employment holds up while only the youth segment slips. Employment among those aged 65 and over rose by 333,000, extending an increase to 66 consecutive months.
Bin Hyun-jun, director general of the Social Statistics Bureau at the National Data Office (국가데이터처), explained that the July employment rate fell for a fourth consecutive month and that unemployment rose among both young people and those aged 30 and over, pushing up the overall unemployment rate. By industry, manufacturing employment fell by 68,000 (-1.6%) for a 25th month, and construction fell by 57,000 (-3.0%) for a 27th month. The change in total employment came down from the 100,000-200,000 range in January-March this year to the 70,000 range in April, then turned to a decline of 40,000 in May, and rose again to 63,000 in June and 108,000 in July.
Explanations that point straight to AI as the cause are increasing. The International Labour Organization (ILO) report "World Employment and Social Outlook 2026," released on the 12th, said AI threatens the jobs of 5.6 million young people worldwide. The same report found that 29.5% of those aged 15-24 who have completed higher education fall within the range at risk of having their work replaced or automated by AI. Of these, high exposure accounts for 11.1% and medium exposure for 18.4%.
It is hard to read the two figures by the same measure. Whether 5.6 million and 29.5% are values drawn from the same population, and whether "threat" means full replacement or a change in part of the work, cannot be sorted out from the report excerpt alone. South Korea's 6.8% (ages 15-29) and the global youth unemployment rate of 12.4% (ages 15-24) compiled by the ILO also cannot be compared directly, since the age brackets and regions differ. The number of unemployed young people worldwide was put at 67 million, and young people not in employment, education or training at 257 million.
High AI exposure does not itself mean job loss. The ILO report found that in medium-exposure occupations a complementary relationship between people and technology is likely to form. A study presented at an NBER conference and introduced by the Bank of Korea draws a similar fork.
Even under extreme conditions in which AI costs fall 90%, use rises ninefold and the data depreciation rate falls 90%, if people and AI maintain a complementary relationship the long-term unemployment rate stays within 6.5% over the next 20 years. If that relationship disappears, it jumps to around 15%.
The cuts actually tallied are smaller than the forecasts. Based on an FMC Group report, job cuts publicly announced in the United States citing AI from 2023 through August this year come to about 175,796. Because it counts only cases in which employers themselves gave AI as the reason, the figure may be lower than the reality, but it is a different order of magnitude from the 85 million jobs the World Economic Forum projected in its 2020 report would disappear by 2025, or the forecast PwC presented in 2017 that 38% of US jobs could be automated.
Dario Amodei, CEO of Anthropic, said in "The Adolescence of Technology" in January this year that within five years 50% of entry-level office jobs would be "affected" by AI. The word he chose is affected. The word "entry-level" narrows the scope.
Among AI experts, the diagnosis is that the danger facing young people is the severing of the path along which they built judgment and expertise by doing the work themselves. The tasks new hires first took on are mostly repetitive and standardized. Organizing materials, drafting, checking figures against one another.
The ILO's pointing to the decline of mid-skill jobs such as office, administrative and sales work points to the same layer. When that layer thins, the road to becoming a third-year professional itself narrows.
Results observed in the area of learning lend weight to this concern. In a Chinese study cited by the head of the AI initiative at the Center for Global Development (CGD), Hanseong Chia (한성 치아), students who submitted homework with AI scored 18% higher on assignments, but their monthly exam scores fell 20% and their entrance exam scores fell 18-24%. A controlled trial in Türkiye also reported that chatbots giving only the answers produced learning losses. It is early to apply the two studies, whose samples and designs have not been confirmed in published form, directly to skill formation in the workplace.
There is no statistic separating out how much of South Korea's 6.8% youth unemployment rate AI explains. For youth employment, down for 45 consecutive months, an accurate diagnosis is possible only by separating the share due to the decline in the youth population itself from the share due to the falling employment rate.
Miguel Sánchez Martínez, an ILO economist, proposes different prescriptions according to income level. High-income countries should give priority to protecting the channels through which young people enter careers and to putting AI retraining systems in place, while middle-income countries should give priority to matching the skills graduates have learned with the capabilities the AI era demands.
The next employment trends report comes out in a month. Rather than whether youth employment fell for a 46th month, this debate will be settled when it is possible to count how many entry-level job postings there were and how many people lasted their first year in those positions. Right now there is no statistic that counts that number.
