As physical AI moves into production sites, what is threatened is less the total number of jobs than the middle step where young workers learned the work and moved on to higher positions, an assessment that has reached the policy stage. Physical AI refers to a form in which artificial intelligence, once confined to language and images, works directly in the physical world through robot arms, legs and sensors. The forum "Jobs and Security in the Age of Physical AI," held on May 20 at the National Assembly Members' Office Building, and the launch on the morning of the 22nd of a labor-management consultative body under the Economic, Social and Labor Council came two days apart.
At the forum, Song Young-hee, visiting professor at aSSIST University and former KT executive vice president, assessed that AI takes away work in the middle zone first. The labor market is splitting into low-skill and high-skill poles, and the "job growth ladder" by which workers stepped through the middle to higher positions has disappeared. She saw the clear preference for experienced hires in the recruitment market as another expression of the same phenomenon.
The figures support this assessment. The "Canaries in the Coal Mine" report from Stanford University's Digital Economy Lab, headed by Erik Brynjolfsson, found that since generative AI spread, employment among people aged 22-25 in occupations with high AI exposure fell about 13%. The decline was concentrated at workplaces that used AI to replace people.
No such decline appeared at places that used it to build up people's capabilities. Korea's youth employment rate stood at 43.3% as of February 2026, down for 22 straight months.
Robots on site are not an abstract future. Boston Dynamics released video on May 19 of its humanoid "Atlas" lifting a 23kg compact refrigerator, and also demonstrated its quadruped robot "Spot" fitted with Google's generative AI "Gemini," taking instructions in human speech. Hyundai Motor Group, at a recent briefing for overseas institutional investors, laid out a plan to bring more than 25,000 Atlas units into Hyundai Motor and Kia production bases. The deployment schedule and the processes involved have not been presented in fixed form, and labor opposition to the plan is spreading.
The way training data is gathered has changed as well. Eom Sung-yong, professor and head of the Department of Knowledge Management at Chung-Ang University, said factory workers in India and Vietnam work while wearing headbands fitted with cameras and sensors. Human work motions are collected as material for teaching physical AI.
Hiring statistics give a sense of where the ladder broke. The Korea Research Institute for Vocational Education and Training analyzed more than 208,000 AI-related job postings from the first half of 2024 to the first half of 2025 in issue 319 of its "KRIVET Issue Brief" on May 21, and found hiring divided into five roles: researcher, engineer, designer, AI+X specialist and citizen. Engineers made up the largest share at 53.0%, but that share was broken into pieces such as model development at 23.9% and data processing and analysis at 19.4%, so as a single job the AI designer led at 36.3%.
AI+X specialists accounted for 10.4% and researchers for 0.3%. KRIVET said designer hiring looks more at actual experience using AI and a record of planning work than at education level or major.
Responses were presented as well. Professor Song proposed giving hiring credit to those who complete micro-degrees, providing entry-level vouchers for high-performance AI computing, and creating a shared-growth solidarity fund for AI new industries. A cross-ministry "National Digital Talent Committee" bringing together talent development programs scattered across the Ministry of Education, the Ministry of Science and ICT and the Ministry of Employment and Labor, and a "K-AI employment model" running from education through entry into practical work, safety nets and the discovery of new industries, were also on the list of proposals. A plan to change the employment insurance standard from working hours to individual total income and to build a one-to-one matching fund between government and companies was also raised.
Other countries have chosen to lay down the first step for young people through institutions. Singapore runs AIAP, in which soon-to-graduate young people carry out industry projects for nine months while drawing a salary, and it also runs SkillsFuture, which grants lifelong learning credits at age 25. Britain has an apprenticeship levy under which companies fund youth training, Estonia teaches coding and AI principles from kindergarten, and Japan has designated high school programming as a required subject. In the United States, big tech leads STEM education and in-house reskilling.
The risks do not rest on employment alone. Yonsei University's Barun ICT Research Center pointed out that the moment robots handle the physical world, the consequences of hacking go beyond data leaks to physical accidents and loss of life, paralysis of infrastructure and military misuse, and proposed making disclosure of robot vulnerabilities mandatory and taking the lead on a Korean international standard for physical AI security.
The "Labor-Management Shared Growth Committee on the AI Transition," which held its first meeting on the 22nd in Jongno-gu, Seoul under the Economic, Social and Labor Council, is led by Hwang Deok-soon, former president of the Korea Labor Institute, and is made up of 17 members: three from labor, three from management, four from government and six from the public interest. Its term of operation is one year. The agenda it has agreed to address runs along four lines: the impact and actual state of AI adoption, labor-management shared-growth use and responses to changes in job duties, acceptance of data collection and use, and building a transition support system. An item aimed squarely at entry-level hiring and the mid-skill path is not among them.
At Davos in January, Anthropic's Dario Amodei warned that a substantial share of white-collar entry-level jobs could disappear within a few years, and Google's Demis Hassabis also acknowledged warning signs for entry-level positions and internships. The International Monetary Fund estimated that about 40% of jobs worldwide could fall within AI's reach, and the International Labour Organization projected that women and office workers would be relatively more affected. There are also counterarguments that humanoids will not fully replace people until after 2050, so differences over timing have not narrowed.
More robots in the factory does not mean the first day of work at twenty disappears. What changes is what that first day leads to three years later. If robots take on those three years of processes in which the hands learned and the eyes opened, young people get work and still lose a place to set a foot on the next rung. The one-year clock of the Labor-Management Shared Growth Committee started running on the 22nd, and whether this issue comes within that clock will become clear in the agenda adjustments of the first few months.
