AI Is Compressing Wages for Millions of US Workers, Apollo Study Finds
New research reveals generative AI is slowing pay growth rather than eliminating jobs, driving a $28 billion annual loss in worker earnings.
Artificial intelligence is driving a subtle restructuring of the labor market marked by stagnant paychecks rather than widespread layoffs. According to a study published on July 30, 2026, by Apollo Global Management chief economist Torsten Slok and co-author Sania Edlich, companies are using productivity gains from generative AI to suppress wage growth in exposed roles rather than downsizing their workforce.
By analyzing real interaction logs from Anthropic’s Claude Economic Index matched with Bureau of Labor Statistics data across 321 occupations from 2015 to 2025, the researchers found that workers in high-AI-exposure roles experienced a 6.7 percentage point deceleration in real wage growth post-2023 compared to unexposed peers.
The empirical study shows that the wage squeeze hits lower-earning demographics hardest. Employees in the lowest wage quartile saw relative wage growth drop by 10.7%, while the second and third quartiles recorded declines of 5.4% and 4.0%, respectively. Conversely, the top wage quartile showed no statistically significant pay erosion, indicating that higher earners possess greater leverage to absorb or capitalize on AI integration.
Across industry classifications, service roles experienced a 24.3% relative decline—though drawn from a smaller sample size—while management and professional occupations recorded a 4.1% reduction. Blue-collar occupations showed no statistically significant wage compression. In total, the researchers estimate that approximately 5.8 million American workers, or roughly 3.7% of the total labor force, are affected, resulting in an estimated $28 billion in annual lost labor income.
These findings represent a notable pivot for Slok, who earlier in 2026 argued that macroeconomic data showed zero evidence of AI-induced disruptions. In April, Slok noted that AI was “everywhere except in the incoming macroeconomic data,” and on May 29 published research invoking the Jevons Paradox—the economic theory that efficiency gains drive higher overall consumption—to argue AI was creating more jobs than it eliminated. Anthropic CEO Dario Amodei similarly referenced the principle after walking back earlier warnings about broad job destruction.
The broader macroeconomic indicators obscure these microeconomic wage shifts. In late July 2026, Anthropic’s head of economics published an analysis of 18 months of internal data, concluding that the overall U.S. labor market remains stable. With the national unemployment rate sitting at 4.2%—a level the Federal Reserve treats as full employment—and prime-age employment near multi-decade highs, macro data appears unaffected. Furthermore, a July 2026 literature review cited by Reuters concluded that AI adoption primarily leads to task reallocation and internal productivity gains rather than widespread job cuts.
Wage suppression tied to technological advance reflects historical patterns observed during previous industrial transitions. During early 19th-century industrialization, textile mechanization lowered hand-loom weavers’ pay long before creating new factory roles, giving rise to the Luddite movement. Similarly, the deployment of manufacturing robotics in the 1980s and 1990s corresponded with decades of stagnant real earnings for industrial workers. Financial Times analyst Joel Suss highlighted that labor’s share of GDP across North America, Europe, and Japan has declined relative to capital since 1970, with AI functioning as capital-biased innovation that widens the pay-productivity divide.
On the ground, economic pressures have fueled widespread workforce pushback against corporate AI strategies. A June 2026 Software Finder survey of 1,005 employed U.S. workers revealed that 50% actively resist new AI tools, with 45% citing fears of becoming replaceable. Workers resisting AI earned an average of $65,645 compared to $81,526 for adopters, reflecting the higher concentration of management roles among early users. Additionally, 13% of surveyed employees admitted to faking AI adoption, while only 6% believed managers understood actual usage levels.
Discontent has also escalated into workplace disruption. An April 2026 survey of 2,400 knowledge workers and 1,200 C-suite executives across the U.S., UK, and Europe by Writer and Workplace Intelligence found that 29% of workers—and 44% of Gen Z respondents—admitted to intentionally sabotaging corporate AI initiatives. Documented actions included uploading sensitive company data into unauthorized public AI platforms, deploying unapproved “shadow AI” software, refusing assigned tools, and altering performance reviews or producing sub-par work to undermine tool efficacy, with 30% citing job security fears as their primary motive.
