Estimated reading time: 4 minutes · Last updated:
Generative AI may be lifting the floor under U.S. joblessness. A Federal Reserve Board paper by Hie Joo Ahn and Nicholas A. Carollo estimates that the natural rate of unemployment has risen by about 0.2 percentage point since large language models (LLMs) arrived. The authors attach considerable uncertainty to that figure. The mechanism is slower movement between jobs: workers with the heaviest AI exposure and adoption have seen job-finding and job-switching rates drop more than other groups. The natural rate is the jobless level an economy settles at once short-term swings fade. The paper was posted to the Federal Reserve Bank of San Francisco's Technology & Growth Center on November 17, 2026.
Key takeaways
- Natural rate: Hie Joo Ahn and Nicholas A. Carollo estimate the U.S. natural rate of unemployment has risen about 0.2 percentage point since LLMs arrived, with considerable uncertainty.
- Job mobility: After LLMs appeared, the rates at which workers found new jobs and moved between jobs fell further for those with high AI exposure and adoption than for any other group.
- Wages and hours: The positive association between AI and nominal wage growth and hours worked weakened sharply in the post-LLM period.
- Since 2023: Dispersion in job-finding prospects has widened markedly since 2023, mostly because of AI factors, according to the authors' hirability index.
- Data: The study combines CPS and JOLTS labor data with an AI exposure measure from OpenAI and an adoption measure from Lightcast.
Table of contents
How the Researchers Measured AI's Reach Into Jobs
Ahn and Carollo paired two standard U.S. labor datasets, CPS and JOLTS, with outside gauges of how much AI touches each worker. OpenAI supplied the exposure measure. Lightcast supplied the adoption measure.
From that mix they rebuilt the labor force group by group. Stocks are head counts at a point in time, such as the number unemployed. Flows are the moves people make between those states. The authors also tracked market tightness for each slice of AI exposure and adoption.
Fewer Job Moves, More Reshuffling Inside Jobs
The clearest change since LLMs appeared is that heavily exposed workers move less. Their job-finding and job-switching rates have fallen further than those of other groups. Meanwhile, how often they shift activities without leaving their employer has climbed noticeably relative to everyone else.
AI once came with a bonus. Its positive link to nominal wage growth and hours worked faded sharply after LLMs arrived. The authors read this as reallocation happening inside companies, as firms reorganize tasks and rework their mix of labor, while demand for the most AI-exposed workers softens.
Why a 0.2 Point Rise in the Natural Rate Matters
To size the pressure, the pair built individual-level hirability and separability indices, which score each person's odds of being hired or of leaving a job. Dispersion in job-finding prospects has widened markedly since 2023, and AI factors drive most of that widening.
The authors recover the natural rate two ways: from trend components of flows into and out of unemployment, split by AI exposure and adoption, and from a theoretical model of structural unemployment. Their result is a rise of about 0.2 percentage point, which they describe as highly uncertain.
| Measure | Change for high-exposure workers | Relative to |
|---|---|---|
| Job-finding rate | Larger decline | Other groups |
| Job-switching rate | Larger decline | Other groups |
| Within-job activity switching | Noticeable increase | Other groups |
| AI effect on nominal wage growth | Weakened sharply | Pre-LLM period |
| AI effect on hours worked | Weakened sharply | Pre-LLM period |
| Dispersion in job-finding prospects | Widened markedly since 2023 | Mostly AI-driven |
| Natural rate of unemployment | Up about 0.2 percentage point | Before LLMs |
What the evidence supports and what it does not
The case for
- Rising within-job activity switching suggests firms are redeploying exposed workers to new tasks rather than only shedding them.
- The estimated rise in the natural rate is about 0.2 percentage point, a modest shift by the authors' own figure.
- Two separate routes, worker-flow trends and a structural model, feed the estimate, giving it more than one footing.
The case against
- Exposed workers' job-finding and job-switching rates have fallen more than other groups', limiting their mobility.
- AI's earlier positive link to nominal wage growth and hours worked has largely faded since LLMs arrived.
- Widening dispersion in job-finding prospects since 2023, mostly from AI factors, points to uneven outcomes across workers.
What to be careful about
- The authors attach considerable uncertainty to the 0.2 percentage point estimate, so the true shift could be smaller or larger.
- Results depend on third-party proxies, OpenAI's exposure measure and Lightcast's adoption measure, which may misclassify some jobs.
- The post-LLM window is short, so trends since 2023 could reflect other forces the indices attribute to AI.
The bottom line
The study's central finding is that AI's labor-market footprint so far shows up in mobility. Workers most exposed to the technology are finding and switching jobs less often, while their roles shift from within. If the natural rate has risen 0.2 percentage point, any given jobless rate would signal slightly less slack than before, which bears on how the Federal Reserve reads the job market. Ahn and Carollo present the figure as early and uncertain. The open question is whether slower job finding for exposed workers is a transition cost that fades or a lasting feature of an economy reorganizing around AI.
What to watch
- Watch for a revised or published version of the Ahn and Carollo paper that could narrow the uncertainty around the 0.2 percentage point estimate; no date has been set.
- Watch whether later CPS and JOLTS data keep showing weaker job finding for workers with high AI exposure and adoption; no specific release has been named.
Frequently asked questions
Has AI raised unemployment in the US?
Hie Joo Ahn and Nicholas A. Carollo estimate the natural rate of unemployment, not the headline rate, has risen about 0.2 percentage point since LLMs arrived. They stress the estimate carries considerable uncertainty.
Which workers are most affected by LLMs?
Workers with high AI exposure, measured with OpenAI data, and high adoption, measured with Lightcast data, saw the biggest drops in job-finding and job-switching rates. Their activity switching within existing jobs rose.
Is AI still boosting wages?
The paper finds AI's positive link to nominal wage growth and hours worked weakened sharply after LLMs appeared. The authors tie this to task reorganization inside firms and softer demand for the most exposed workers.
Related reading