AI May Be Squeezing Workers' Pay Before It Cuts Their Jobs

Key points

• Workers in jobs rated highly exposed to AI saw real-wage growth run 6.7 percentage points slower after 2023 than workers in less-exposed jobs, according to an Apollo Global Management study.
• The same study found no statistically significant effect on employment, suggesting pay compression may arrive before layoffs do.
• Labor's share of US nonfarm business income hit 52.8% in Q2 2026, the lowest recorded since tracking began in Q1 1947, according to the Bureau of Labor Statistics.
• The Apollo study covered only 321 of roughly 800 BLS job categories, and only 11 met its high-exposure threshold, limiting how much weight the findings can carry.
• MIT economist Daron Acemoglu expects wage effects to ultimately outrun job-loss effects in the US, given the country's flexible labor market and limited worker protections.

Hiring beat expectations in August, unemployment stayed low, and the mass layoffs that AI doomsayers predicted haven't arrived. But a quieter signal is drawing economists' attention: wages, adjusted for inflation, are softening in exactly the occupations where AI tools are most widely used.

A study from Apollo Global Management's chief economist Torsten Slok and co-author Sania Edlich put a number on it. Workers in occupations the researchers classified as highly exposed to AI, meaning AI tools can handle a meaningful share of those workers' tasks, saw their real wages grow 6.7 percentage points more slowly after 2023 than workers in less-affected jobs. Employment in those same roles showed no statistically significant change. The authors suggest companies may be capturing productivity gains from AI not by cutting headcount but by holding pay down.

That framing matters. If AI reduces what a worker is worth to an employer before it actually replaces that worker, the harm shows up in a paycheck long before it shows up in an unemployment figure. We looked at weakening employment figures from a different angle in our 13 August story "US Economy Lost 23,000 Jobs in July. Should We Blame AI?".

How strong is the evidence?

Not strong enough to draw firm conclusions, and the researchers say so themselves. The Apollo study is described by its own authors as "early evidence" built on a narrow data set. Only 321 of roughly 800 Bureau of Labor Statistics job categories could be included, and just 11 cleared the bar for high AI exposure. That's a small sample.

Ben Zipperer, senior economist at the Economic Policy Institute, a left-leaning research group, acknowledges AI could be pushing wages down in certain roles. But he flags a measurement trap: if AI makes software cheaper to build, the money companies save doesn't vanish. It gets spent elsewhere, lifting wages in other parts of the economy. Comparing highly exposed jobs against that rising baseline makes exposed jobs look worse than the underlying reality may warrant. Zipperer also notes that the recent slide in tech wages partly reflects companies unwinding the over-hiring of 2021 and 2022, a correction with no connection to AI.

Daron Acemoglu, professor of economics at MIT, adds a structural caution. AI tools are still concentrated in a narrow band of tasks and industries, so displacement effects measured now may be overstated. He sees "mounting evidence" of impact on entry-level jobs, and he expects, given how loosely the US labour market protects workers, that wages will absorb more of the shock than employment levels will. "given that the U.S. Labor market is relatively flexible and has a fairly weak social safety net, I expect the impact on wages to be bigger than those on employment," he told CNBC.

What should workers make of this?

Watch your pay, not just your job security. The broader BLS figures already show inflation-adjusted wages and salaries fell 0.4% year over year through June 2026. Labour's share of business income is at a record low going back to 1947. Direction is consistent; confidence in the cause is not.

MIT labor economist David Autor cautions against reading "AI exposure" as any kind of verdict on a job's future. His research on accounting clerks versus inventory clerks, two roles that looked equally vulnerable to automation decades ago, found they ended up on opposite trajectories. Accounting clerks saw wages rise 39% over 40 years even as employment fell; inventory clerks saw wages drop 13% even as the number of jobs grew 175%. The technology touched both. Outcomes split entirely on whether AI absorbed the expert parts of the work or the routine parts.

That's the detail most coverage skips. Knowing a job is exposed to AI tells you it'll be affected. It tells you almost nothing about whether the person doing it ends up better or worse off, and right now, neither does the data.