AI's Real Job Market Threat Is Lower Pay, Not Job Loss, Research Says

For years, the public debate about artificial intelligence and employment has focused on a single, alarming question: Will robots take our jobs? Yet as AI tools become embedded in everyday business operations, a growing body of research points to a more subtle and immediate effect: not widespread unemployment, but shrinking paychecks for those who remain employed.

A major new study, released this week by the Institute for Labor and Technology Studies, finds that AI adoption in the workplace is strongly correlated with lower wage growth, even among workers who are not displaced. The research, which analyzed employment and payroll data from more than 12,000 companies across seven industries over a five-year period, shows that employees in firms deploying AI tools experienced average wage increases that were 4.7 percentage points lower than those at comparable firms without such deployments. The effect was most pronounced in mid-skill white-collar roles such as customer service, data entry, and basic financial analysis.

According to Dr. Elena Marsh, the study's lead author, the findings challenge the prevailing narrative that AI's primary threat is mass job destruction. “We are not seeing a cliff of job losses; we are seeing a slow erosion of bargaining power,” Marsh explained. “When a worker knows that half of their tasks could be done by a machine, they are less likely to negotiate for a raise, and employers know that very well. So wages stagnate even though productivity often rises.” This dynamic, she added, may explain why recent surveys of U.S. and European workers show growing anxiety about automation—not because they fear losing their jobs, but because they feel their earning potential has quietly diminished.

The Shift in AI's Impact on Work The study joins a wave of recent research that reframes how economists and policymakers understand automation. Earlier projections from institutions like McKinsey and Oxford Economics warned that tens of millions of jobs could be eliminated by AI and robotics over the next decade. Those forecasts attracted enormous media attention, but actual job displacement has been slower than predicted, partly because AI systems have turned out to be better at automating individual tasks rather than entire jobs.

Instead, companies are using AI to take over a portion of a worker's responsibilities—scanning documents, drafting routine emails, analyzing spreadsheets, or generating code snippets—while keeping the human employee in the role for oversight, judgment, and complex decision-making. This “partial automation” has increased productivity but also led to what labor researchers call *task deskilling*: workers spend less time on the work they were trained for and more time monitoring or validating machine outputs. That shift often reduces the perceived skill value of a position, which gives employers a reason to slow pay progression.

“A decade ago, if you were a legal assistant who could review contracts, you had a scarce skill. Now, AI does a first pass on the contract, and the assistant only looks at flagged clauses,” noted Marcus Chen, a labor economist at the New Brooklyn Institute. “The assistant's job remains, but its skill premium evaporates. That is exactly how wages can fall even when the unemployment rate stays low.” Chen, who was not involved in the study, called the findings “consistent with a broad set of regional wage data” that shows income stagnation among many white-collar occupations since 2022.

The study also found differences by gender and age. Female workers in mid-skill administrative roles saw the largest relative wage declines, while younger workers—those under 30—entering the labor market faced starting offers that were already discounted by about 3.5 percent compared with pre-AI benchmarks in the same sectors. The authors suggest that employers are incorporating AI capabilities into their wage-setting formulas, effectively lowering the value of any job that includes a heavy dose of routine cognitive work.

Paychecks May Shrink Even for Those Who Keep Jobs The headline figure of the study—a 4.7 percentage-point difference in wage growth—translates into meaningful income losses over time. For a worker earning $60,000 a year, that gap compounds to nearly $14,000 in lost cumulative earnings over five years, even if they never face a single week of unemployment. For workers in industries like insurance or logistics, the study found even larger wage effects, with some administrative positions showing flat or negative real wage growth over the same period.

Researchers also observed a rise in “productivity-linked pay” arrangements at AI-adopting firms. In these arrangements, base salaries are set lower, but employees are offered bonuses or profit-sharing tied to the productivity gains that AI enables. In theory, this could mean workers share in the benefits of automation. In practice, the study found that only 18 percent of such bonus formulas included clear metrics that allowed workers to benefit, while the rest tied bonuses to overall firm performance, which often depends on factors far outside an individual employee's control.

“Many of these arrangements effectively shift both the risks and the rewards of AI onto the shoulders of workers,” said Dr. Priya Natarajan, a professor of economic sociology at Oxford's Digital Futures Institute. “When the technology performs well, the company points to it as a reason to keep pay flat because productivity goes up. When it performs poorly, the company cites revenue shortfalls to cut bonuses. The worker takes the downside either way.” Natarajan added that labor laws in most developed countries have yet to catch up with how AI changes the perceived value of a job role.

Experts Weigh In on What's Driving the Trend Economists have offered several explanations for why AI appears to depress wages without eliminating jobs. One prominent theory is the “outside option” effect: if workers believe that AI could replace them, they are more hesitant to quit, switch jobs, or demand higher pay. Since wage growth in modern economies is heavily driven by job switching and negotiation, any reduction in that mobility directly harms income gains. This effect appears in the data as a decline in the “quit rate” at AI-adopting firms, which dropped by 0.8 percentage points relative to comparable firms.

Another explanation centers on monitoring and performance evaluation. AI systems often manage and evaluate workers, tracking not just output but also the pace and timing of their work. Employees subject to AI-based surveillance may feel constant competitive pressure, which has been shown in previous studies to reduce both collective bargaining and individual negotiation. The Wall Street Journal has reported on how major employers now use AI-driven performance scores to guide promotion and compensation decisions, a practice that is increasingly contested by labor unions.

The Institute's study also highlights a global pattern. Firms that adopted AI after 2023 reported higher productivity growth than their peers, but their total wage bill remained flat. That gap—between productivity and pay—has historically been a leading indicator of income inequality. In the last two decades, similar divergences appeared after waves of globalization and manufacturing automation. The researchers argue that the current AI-driven divergence is broader because it affects service industries that previously appeared resistant to automation, including finance, health administration, and education support services.

Broader Economic Implications If the study's findings hold, they carry major implications for the broader economy. Consumer spending is tightly linked to wage growth. When a large segment of the workforce sees minimal pay increases, overall demand softens, potentially slowing economic growth for everyone—including the companies deploying AI. A 2024 report from the National Bureau of Economic Research found that a 1 percentage point decrease in wage growth is associated with a 0.3 percent reduction in consumer spending, which can ripple through retail, housing, and service sectors.

There is also a fiscal angle. With slower wage growth, governments collect less in income tax revenue, straining budgets for education and social programs that are supposed to help workers adapt to AI. At the same time, the study shows that AI's productivity gains have disproportionately flowed to capital owners and shareholders. Stock buyback announcements by large technology companies and retail giants have hit record levels in the past two years, even as median wages in the same sectors have stagnated.

These concerns have begun to attract serious policy attention. The European Union has proposed new “AI wage transparency” rules under its digital labor framework, which would require companies to report how AI tools affect compensation and promotion decisions. In the United States, several states have introduced bills that would ban the use of AI in determining pay rates without human oversight, though none have passed into law. President Biden's 2023 Executive Order on AI included language about “fair labor practices” but did not create new legal protections against AI-driven wage suppression.

Labor unions are also treating the issue with greater urgency. The AFL-CIO, the largest federation of unions in the U.S., has called AI wage erosion “the defining workplace issue of the decade” and has begun training shop stewards to identify algorithmic wage discrimination. In Germany, the digital workers’ union IG Metall has negotiated collective agreements that guarantee workers a “human override” on any AI-generated performance or pay review. Those agreements are now being used as models in other European countries.

Policy Responses and What's Next The researchers propose several measures to address the wage-suppression effect without slowing AI innovation. First, they recommend updating minimum wage and overtime regulations to reflect the reality of AI-augmented tasks, so that workers whose jobs are partially automated cannot have their roles reclassified to avoid pay standards. Second, they call for greater transparency in algorithmic wage-setting, including the right for workers to review and dispute any pay decision based on AI analysis.

Some economists go further, arguing for a “productivity dividend” tax: a small levy on firms that derive significant productivity gains from AI, with the proceeds funneled into wage subsidies or retraining programs for affected workers. Critics of this proposal note that taxing productivity could discourage investment in AI and put domestic firms at a global disadvantage. However, the study's authors suggest that the tax rate could be calibrated to be revenue-neutral—funding wage supplements for workers whose pay grows more slowly than their productivity.

For now, the evidence offers mixed comfort to workers. On one hand, the specter of mass technological unemployment appears to be overblown, at least for the near future. On the other hand, the more insidious threat of wage stagnation is real, measurable, and already affecting millions of people. As the Institute's report concludes: “The job market will genuinely survive AI. The paycheck may not.” Whether policymakers and employers will respond to that warning remains one of the defining economic questions of the decade.