Hiring Freezes: The Overlooked Side of AI and Jobs

A Question That Starts From an Expert Survey

In a survey reported by Seoul Economic Daily (Sedaily) on September 6, 2026, 82.7% of respondents said the pace of AI adoption shouldn’t be slowed down. 74.0% also said that high-risk personnel decisions, like dismissals, need control mechanisms.

The survey covered 102 labor experts—corporate HR executives and professors among them. When reading these results, we need to consider the scope of who was surveyed. We can’t treat expert opinion as equivalent to the views of the entire workforce.

The two responses can coexist. They can be read as saying: use AI, but handle decisions that affect people’s jobs carefully.

I want to add one more question here. When we look at the impact on employment, are we considering both the dismissal of people already working and the hiring plans for people who haven’t been hired yet?

Using AI for Work Is Different From Using It for Personnel Decisions

In the Bank of Korea’s 2025 household survey, 63.5% of workers said they use generative AI. When limited to work purposes, the usage rate was 51.8%. I covered this in Issue 206.

This figure shows that AI has widely entered individual work. But it doesn’t tell us whether people used it without company approval, or whether it’s also used in hiring and evaluation.

An individual summarizing documents and a company screening applicants are different in both the subject of the decision and who bears responsibility for it. The fact that employees already use AI doesn’t mean the organization no longer gets to set the terms of adoption. What data to allow, how far to apply it in personnel work, and who reviews it—these all still need to be decided.

So it’s hard to say the debate over the pace of adoption is settled. What remains is which tasks get done with what authority.

What the Statistics on Shrinking Hiring Plans Actually Show

According to the Ministry of Employment and Labor’s Job-Type Business Labor Force Survey for the first half of 2026, the labor shortage stood at 467,000 as of April 1—the number of workers businesses said they needed for normal operations. In the same survey, hiring plans for the second and third quarters came to 460,000, down 1.8% from the same period a year earlier.

These two figures make you think about the gap between needed headcount and actual hiring plans. But the shortage at a single point in time and hiring plans for the next two quarters are different metrics. You can’t take the difference between the two totals and read it as the number of positions companies deliberately left open.

In the same release, actual hiring in the first quarter came to 1,368,000, up 4.6% from a year earlier. Actual hiring performance and subsequent hiring plans moved in different directions. So this data can’t be used as evidence that “hiring at Korean companies has fallen across the board.”

Among reasons for unfilled positions, the most common—at 25.8%—was a lack of applicants with the required experience. That means required experience is one obstacle to hiring; it doesn’t mean every company has stopped hiring entry-level workers.

Nor can this statistic alone tell us that AI is the cause of the decline in hiring plans. We also need to look at the business cycle and outlook, labor costs, and the gap between what jobseekers and companies each expect.

We Need to Look at Both Youth Hiring and What Happens After

There’s also a Bank of Korea study from August 2026 that directly examines AI and youth employment. It’s an analysis using National Pension Service enrollee data and the Economically Active Population Survey.

The study finds that youth employment declined notably in industries with high AI exposure. But alongside the drop in new hiring, it also observed existing young workers leaving their jobs. That’s different from a picture in which employment changes happen only by cutting new hiring.

There were also differences depending on how AI was used. Youth employment fell sharply in industries where AI mainly automates tasks, while the same pattern didn’t appear in industries where AI mostly assists human work.

The Bank of Korea cautions against concluding this is purely a direct effect of AI, since post-pandemic corrections to over-hiring, a preference for experienced hires, and changes in in-house training could all have played a role together.

What I want to focus on here is the opportunity for young people to start working and build experience. I think we need to look not just at current headcounts, but at who’s coming in and how much they get to learn on the job once they do.

How Should We Manage AI in Personnel Decisions?

Article 2 of the AI Basic Act (Korea’s framework law on artificial intelligence) defines high-impact AI’s scope of use to include judgments or evaluations that significantly affect an individual’s rights and obligations—hiring and loan screening among them. The mere fact that the word “dismissal” isn’t spelled out doesn’t mean AI used in dismissal decisions is always excluded.

Article 34 requires AI businesses that provide such AI, or products and services using it, to take measures such as risk management, explanation plans, user protection, human oversight, and drafting and retaining related documents. Article 27 of the Enforcement Decree requires that the documents evidencing compliance be kept for 5 years.

We shouldn’t read human oversight here as meaning an identical form of final approval requirement applies to every personnel decision. Whether a given system counts as high-impact AI, and who bears which obligation, has to be examined case by case, depending on the service and the role involved. The responsibilities of the company supplying the AI and the company using it also need to be distinguished.

The fact that the law imposes a management obligation is separate from an assessment of whether explanation and oversight are actually adequate in practice. You need to check what criteria were applied to an applicant, who re-reviews the case when an error is found, and whether you can obtain the necessary materials from the vendor.

Oswarld’s Lens

What worries me is that we only discuss the decisions that are easy to see. A dismissal is announced to the person involved, and a process follows. But a decision to cut hiring plans or delay filling a position is hard for outsiders to notice.

For example, a team’s request to add headcount could be put on hold simply because AI can now handle some of the work. This is one possible decision among others—the statistics cited above don’t measure how many companies actually made that call.

Still, I think organizations need to check. The outcome differs depending on whether hiring was held back because workload actually fell, because of cost cutting, or because existing employees took on more work. And if entry-level positions where people learn the job have shrunk, we also need to think about how to develop the experienced hires we’ll need down the road.

If you work in HR or legal, you need to examine both the decision criteria and the review process for the AI used in hiring and in dismissal alike. Setting up safeguards on one side doesn’t mean you can skip looking at the other.

To executives, I’d suggest gathering up all the decisions to hold back on hiring and looking at them together—checking which departments changed their plans and why, and how that changed the workload of remaining staff and the training opportunities for new hires.

When we discuss AI and jobs, dismissal counts alone aren’t enough. We can only see what’s actually changing by looking at new hiring, the departure of existing workers, and workload and training opportunities together. I hope that discussion also makes room for the opportunities of people who haven’t gotten in the door yet.

---

References & Further Reading

• Seoul Economic Daily (Sedaily) report on the survey about AI adoption and control mechanisms for dismissals and the accompanying explanation of the survey’s sample — reported September 6, 2026.
• Ministry of Employment and Labor, Job-Type Business Labor Force Survey, first half of 2026 — lets you distinguish actual hiring, labor shortages, and future hiring plans.
• Bank of Korea, The Rapid Spread of AI and Its Productivity Effects — a 2025 survey of worker AI use.
• Bank of Korea, Study on the Contraction of Youth Employment and Career Formation — August 18, 2026. Covers causes beyond AI along with hiring and attrition.
• AI Basic Act, Article 2, Article 34, Enforcement Decree Article 27 — define high-impact AI and the management and record-keeping obligations of AI businesses.

The author is Oswarld (Kwangseob Ahn). Current roles: Adjunct Professor at Sejong University, Strategy Consultant at INLEVEL9. Career, research, books, and recent work are kept current on the About page. Latest · July 2026: HEMA-2: A Consolidation-Aware Tri-Memory Architecture with Multi-Channel Scheduling for Lifelong Conversational AI.