Executive Summary

Key figures

The first two numbers come from a place that built a field for technology and collected nothing in it. The second two come from the part of the labor market where the statistics never existed to begin with.

Sources: Hunton, Bank of Korea Issue Note No. 2026-19, Aju Business Daily, The Segye Times.

The Green Paper Deliberation Body for a New Social Contract in the AI Era opened on September 9 at the Koreana Hotel in central Seoul. Kang Sung-jin, a professor of economics at Korea University and president of the Korean Economic Association, chairs it, and there are 23 members in all. They include the two national union federations, the Federation of Korean Trade Unions and the Korean Confederation of Trade Unions; the business groups Korea Enterprises Federation and the Federation of Korean Industries; organizations that speak for young, non-regular and unorganized workers, among them Pulppang, the Youth Community Union and the Korean Network of Non-regular Workers' Organizations; and 12 academics drawn from economics, business, political science, sociology and labor law. The body has met every other week since September.

Six themes sit on its agenda: how AI and technological change are reshaping youth employment; new forms of work such as platform, dependent-contractor and freelance labor, and who represents them; the social safety net and the government's role; industrial competitiveness and polarization; how the gains from AI get distributed; and a new model for social dialogue. Reporting at the launch traced the body back to a dispute at Samsung Electronics in May over special bonuses tied to the AI and semiconductor boom. Minister of Employment and Labor Kim Young-hoon called that dispute the starting point for a debate on redistributing excess profits in the AI and semiconductor era.

An earlier document had already set out the chair this body now sits in. The Basic Plan for Employment Stability in Industrial Transition, released on July 9, handed the questions of income gaps and the distribution of gains over to social dialogue, and announced a discussion of a new social contract alongside them. Excess profit redistribution, the original starting point, receded as the windfall tax revenue from the semiconductor boom moved into a future response fund, and the scope of the discussion widened by that much. That last point is how November's reporting describes it.

On November 12 the ministry added a step. Starting as early as the following week, it will spend about three weeks taking proposals directly from the public on what the green paper should cover, and it will fold what it receives into the document. At the September launch Kim said a social contract fitted to the AI era had become necessary, and in the opening address argued that competitiveness in the AI era should not stop at a handful of large companies, but should carry through to workers, suppliers and the innovation capacity of the whole industrial ecosystem. Reporting on the November announcement carried both lines again as the purpose of the call for submissions.

When a company lets a large number of people go for business reasons, the Labor Standards Act requires it to file with the Minister of Employment and Labor. The filing has four fields: the reason for the dismissal, the number of people to be dismissed, what was discussed with the workers' representative, and the timetable. The reason field usually gets something like a business downturn or a contraction of operations. Nothing in the form separates a contraction that followed the arrival of technology from one that did not. There is no box for it.

So the fact that a company cut jobs because of AI disappears the moment the filing is accepted. The diagram below puts the questions the form asks next to the one it leaves out.

▲ Pebblous original diagram. The four boxes reproduce the items required for a business-reasons dismissal filing under the Labor Standards Act.

A bill to fill that blank already exists. The amendments to the Labor Standards Act and the Framework Act on Employment Policy that Representative Yoon Jong-gun introduced on October 21 would require an employer to report to the minister, where a dismissal is wholly or substantially attributable to the adoption or expansion of AI, which tasks are being automated or replaced, how many people in each occupation, and what type of technology is involved. An employer that does not report, or that reports falsely, would face a fine of up to 5 million won.

The bill goes further than creating the field. The minister would have to publish the filings as quarterly statistics broken out by industry, firm size, region, occupation and type of technology, and to report to the National Assembly once a year. The companion amendment to the Framework Act on Employment Policy names AI adoption as a cause within the large-scale employment change filing, which also reaches voluntary and honorary retirement. Redundancy rules in Korea are demanding enough that companies often cut headcount through voluntary retirement instead, so amending only the Labor Standards Act would let those cases slip past. Yoon said at the introduction that nobody in Korea can currently tell whose jobs AI is taking over, or how many.

2.1One year of filings from the state that added the field

Does building the field make the records appear? New York State ran that experiment a year ahead. In March 2025 the state Department of Labor added an option to its mass layoff filing system asking whether technological innovation or automation was among the causes of the layoff, and selecting it brings up a second question about which technology. New York was the first state in the country to put the item on the form. The federal WARN Act does not ask for the reason at all, and since October 1, 2026 Connecticut has required the same question on its mass layoff notices, so New York is no longer the only case.

A year later the count stood at zero. Some 162 employers filed more than 750 notices covering over 28,000 affected workers, and not one of them selected AI or automation as the cause. Over the same stretch Challenger, Gray & Christmas counted at least 48,400 US job cuts attributed to AI or automation during 2025. Those two numbers are hard to read as descriptions of the same world.

California designed it differently. SB 951, signed by Governor Newsom on September 30 and taking effect on January 1, 2027, requires a notice of technological displacement at the top of the document whenever a layoff, relocation or closure is wholly or substantially attributable to AI or automation technology. The employer has to list the number of people and the occupations affected, the worksite, the tasks being automated and the type of technology, and to identify whoever developed, sold or leased it. The larger difference comes next. The state Employment Development Department has to gather these notices and publish them quarterly, and to report to the legislature by January 1, 2028 on how AI has affected hiring across the state and by region. One statute creates the field and orders the counting of what lands in it.

The Bank of Korea issue note published on August 18, asking whether AI is to blame for the youth employment squeeze and subtitled with the changing career ladder and the policy responses to it, examined the four years from June 2022 to June 2026. Jobs held by people aged 15 to 29 fell by 285,000, and 268,000 of those sat in AI-exposed industries. That is a contribution of 94 percent. The researchers classified the top half of industries, ranked by how closely their tasks relate to AI technology, as highly exposed.

Broken out by industry, the decline piles up on one side. The table below gives the four industries the report names.

▲ Rate of change in youth employment by industry, as given in Bank of Korea Issue Note No. 2026-19 (June 2022 to June 2026).

Employment among people in their fifties rose in those same industries. Of the 230,000 jobs added for that age group over the four years, 173,000 came from AI-exposed industries. The report explains that AI takes over manual, rule-following work more readily than the work of experienced staff, so early adoption lands on young workers first. It attaches a caveat as well. The recent weakness in youth employment cannot be laid entirely on AI. The report lists the unwinding of pandemic-era over-hiring, a hiring culture that prefers experienced recruits taken on as needed, and the loss of informal on-the-job learning as remote work spread.

The way in and the way out were counted separately. Measured against the 2016-to-2019 average, the number of young people entering AI-exposed industries each month fell from 32,600 to 29,100, while the number moving from regular employment into non-employment rose from 3,700 to 4,900. The fall on the entry side is larger than the rise on the exit side, and the report describes both routes as working together. Oh Sam-il, who heads the central bank's employment research team, judged it reasonable to read AI as having accelerated a career ladder that was already coming apart.

The limits of the filing surface again here. A dismissal filing is an instrument attached to the door people leave by. The change in youth employment runs closer to shrinking the number of positions people could arrive at. A decision to post one fewer job opening reaches no form at all. The exit side is no different, since most of the moves from regular employment into non-employment were never subject to a business-reasons dismissal filing in the first place. A field for technology on that filing would leave both flows untouched. The 42.9 percent employment rate for 15-to-29-year-olds in the October figures from the Ministry of Data and Statistics, down for the thirtieth month running, is likewise a number collected along a different route from the filings.

The government has not left the counting alone. The July plan from Section 1 carries a measurement track. Rather than borrowing foreign indices, it commits to building an AI exposure index out of Korean occupational data by 2027, and on top of that index to opening a canary dashboard, as early as the second half of 2027, showing employment change by industry and age for the occupations most exposed. The name follows a tool of the same name from the Stanford Digital Economy Lab. That instrument measures how exposed an occupation is to AI. The cause of any individual dismissal stays unknown unless the employer writes it down. The gap Yoon pointed to in introducing the bill sits exactly there.

Employment counts and employment insurance enrollment are built around wage workers. People who take jobs through platforms, freelancers and dependent contractors largely fall outside. A survey presented by Kim Jong-jin of the Working Citizens Research Institute at the ministry's Young Freelancers Open Mic, held on October 15 at the Seoul Regional Employment and Labor Office, shows the size of that gap. Of 462 people aged 19 to 34, 17.5 percent were enrolled in employment insurance. Among the 169 with three or more years of experience, 53.3 percent said they had been treated unfairly, through demands for work outside the contract or cuts to their pay, and more than half worked with no written contract.

The total is harder to measure still. The Ministry of Employment and Labor and the Korea Employment Information Service published a platform worker survey every year from 2021 through 2024. They applied for national statistics approval to put publication on a regular footing, and the application failed. The population frame was too small and not representative enough. So the official number stops at 2023. The 883,000 published in August 2024 is the last one.

▲ Pebblous original diagram. It lays out the publication, gap and planned resumption of the platform worker survey.

A study commissioned by the ministry's dependent contractor support division at the end of 2025 put dependent contractors at about 1.26 million, platform workers at about 800,000 and freelancers at about 660,000, and after removing overlap between occupations estimated the whole at about 2.1 million. The scope differs from the published 883,000, so the two cannot be set side by side. Even so, official statistics and a commissioned estimate put the population a single ministry deals with at very different sizes. The scope has not been settled yet.

The net itself is being rewoven outside the statistics. The employment insurance threshold, 15 hours a week since 1995, changes in January 2027 to monthly pay of 800,000 won. The Korea Workers' Compensation and Welfare Service is building the systems to link National Tax Service income records with employment insurance records. Once income is the test, people kept out because their hours could not be counted come inside the record. The 2027 start applies first to workers under the Labor Standards Act, and extending it to platform workers and freelancers has been deferred, since it depends on how much of their income is reported.

News like this usually reads as labor policy. From where data work sits, something else stands out. A green paper comes one step before a measurement system is designed for a phenomenon nobody has defined yet. Choosing which issues to write down becomes choosing which items to count. Questions first and statistical design after is the same order as agreeing on what counts as a fact before a schema is fixed in data quality work.

New York also shows where that design breaks. The field is one thing, and getting it filled is another. With the scope of what has to be reported undefined, no penalty when it is left empty, and nobody named to collect and publish, the records never accumulate. An optional field that stays permanently empty in corporate data governance has the same structure. Three things sit alongside the questions the green paper will put.

• What would have to be counted anew to answer the question. A question the statistics already answer and a question that needs a new count built first are on different timelines.
• Whether a collector and a publication cycle are named. California and the Yoon bill put the field and quarterly publication in the same statute, while New York changed only the form.
• Who the denominator is. Statistics that count only wage workers make the same event a different size once platform and freelance workers belong in the picture.

Pebblous treats the presence of a field to record a fact as a quality item when it prepares data for AI use, for the same reason. Wrong values are not the only quality problem. An item left empty because there was never a place to record it is harder to find than a wrong value, because an absent value does not register as an error.

When the three weeks of submissions open, one question a reader can send runs like this. Which statistic is currently picking up the change that happened to my job? Whether that question makes the list the green paper publishes at year-end decides what will exist as a number next year.

Official Documents

• 1.Bank of Korea Employment Analysis Team. (2026). "Is AI Behind the Youth Employment Slump? The Changing Career Ladder and Policy Challenges." Bank of Korea Issue Note No. 2026-19.
• 2.California State Senate (Sen. Eloise Gómez Reyes). (2026). "SB 951 — Employment: technological displacement: notice." California Legislative Information.

Korean News Coverage

• 3.Yang, G. (2026). "Ministry of Labor Seeks Public Opinion for the "Green Paper on a New Social Contract in the AI Era"." Newspim.
• 4.Kim, N. (2026). "Finding a Way to Distribute "Excess AI-Era Profits"… Green Paper Council Launches." Kyunghyang Shinmun.
• 5.Kim, Y. (2026). "Jobs and the Safety Net That AI Changed… 23 Labor, Management, and Youth Representatives Discuss a "New Social Contract"." Herald Business.
• 6.Woo, T. (2026). "[Reading Hiring Data] How Does the Mandatory AI Layoff Disclosure Bill Differ From California's SB 951?." Korea Data Economy News.
• 7.Lee, J. (2026). "National Statistics on "Platform Labor" to Launch From 2027." Segye Ilbo.
• 8.Kim, S. (2026). "Young Freelancers' Employment Insurance Enrollment Rate Is 17.5%… Ministry to Enact "Working People's Act" This Year." Ajunews.

Foreign Legal Analysis

• 9.Quackenboss, R. T., Meyer, M. (2026). "New York WARN Act: No AI-Related Layoffs Reported in First Year of Adding AI-Related Disclosure to the System." Hunton Andrews Kurth.