The AI Anxiety Paradox: Workers Most Fearful of Job Loss Are Upskilling, Demanding Raises and Spending More

For years, the dominant narrative around artificial intelligence has been one of looming displacement. Headlines have warned of mass job losses, with estimates ranging from Goldman Sachs’ projection of up to 300 million full-time equivalent roles exposed globally to the IMF’s assessment that roughly 40% of jobs worldwide — and 60% in advanced economies — are exposed to AI. Yet new behavioral evidence suggests the workforce is responding in ways that conventional economic models did not fully anticipate.

A comprehensive analysis by Deutsche Bank Research, drawing on its proprietary dbDataInsights household survey across Germany, France, Italy, Spain, the United Kingdom and the United States, reveals what it terms the “AI Anxiety Paradox.” Workers who express the greatest concern about losing their jobs to AI within the next year are not retreating into precautionary savings or reduced consumption. Instead, they are taking concrete steps to adapt: pursuing AI-related training, demanding higher wages and, counterintuitively, reporting higher spending relative to the previous year.

Informed Concern, Not Irrational Fear

The survey, which has tracked responses since June 2025 using a 0–10 scale for concern about needing to find a new job due to AI, shows stable but elevated levels of anxiety. Crucially, this concern correlates strongly with knowledge and exposure rather than ignorance. Respondents with higher self-reported understanding of AI, greater educational attainment and recent use of AI tools at work or home report significantly higher concern scores.

This pattern holds across age groups, with younger and more educated workers expressing greater unease. Far from being a sign of Luddite resistance, the data indicate that those closest to the technology are most attuned to both its capabilities and its potential to reshape tasks.

The Behavioral Response: Upskilling and Bargaining

The most striking findings concern action rather than attitude. Workers with higher AI-related job loss concern are substantially more likely to have undertaken some form of AI training — watching instructional videos, completing courses or reading relevant material. In regression analysis controlling for demographics, income and country-month effects, a one-point increase in concern on the 0–10 scale is associated with a 0.5 to 1.7 percentage point rise in the probability of upskilling, depending on the subsample.

Even more counter to textbook predictions is the wage response. Standard theory suggests that a technology threatening to substitute for labor should weaken workers’ bargaining power. The survey data show the opposite. Respondents more concerned about AI displacement report a higher likelihood of asking their employer for a pay rise in the next three months. Panel OLS estimates indicate that a one-point rise in concern correlates with a 0.29–0.30 point increase on the 0–10 likelihood scale; ordered logit models show 21–22% higher odds of reporting greater likelihood of seeking a raise.

These patterns persist after extensive controls, suggesting the relationship is not merely compositional.

Spending Behavior Challenges Precautionary Savings Narrative

Consumption-smoothing models predict that heightened job insecurity should prompt households to cut spending and build buffers. The Deutsche Bank data present a more nuanced picture. While concerned workers do report stronger intentions to increase savings over the next 12 months — consistent with precautionary motives — they simultaneously report spending more than they did a year earlier. Ordered logit regressions link each additional point of concern to a 3–6% higher likelihood of being in a higher spending category relative to the prior year.

This combination — maintained or increased spending alongside elevated savings intentions — implies that, at least so far, AI-related anxiety has not produced the broad demand drag some macroeconomists feared. Businesses investing heavily in AI infrastructure may therefore face less immediate offset from household retrenchment than the most pessimistic scenarios suggested.

Exposure Is High, But Realized Displacement Remains Modest

These behavioral findings gain context when set against broader exposure estimates. The IMF’s widely cited figure of 40% global job exposure (60% in advanced economies) measures potential task substitutability, not inevitable job loss. Goldman Sachs updated its analysis in 2025 to suggest that, under current adoption trajectories, only around 2.5% of U.S. employment faces immediate displacement risk from existing AI use cases, with a baseline scenario of 6–7% under wider diffusion.

Early empirical studies of realized effects paint a similarly differentiated picture. Research using high-frequency ADP payroll data (Brynjolfsson, Chandar and Chen, 2025) finds that employment for workers aged 22–25 in highly AI-exposed occupations has declined roughly 16% relative to trend since the widespread adoption of generative AI tools, while employment for older workers in the same occupations has been more stable. Anthropic’s analysis similarly detects no systematic rise in unemployment rates for high-exposure occupations overall, though it notes suggestive evidence of slower hiring for younger workers in those roles.

These patterns align with a task-based rather than occupation-based view of technological change, long emphasized by MIT economist David Autor. Jobs are bundles of tasks. AI automates certain tasks — particularly those involving information processing, pattern recognition and routine cognitive work — but rarely eliminates the entire bundle. The remaining tasks, especially those requiring judgment, context, client relationships or physical presence, often become more valuable. In some cases, AI lowers the expertise threshold for certain activities, potentially expanding the set of workers who can perform higher-leverage work.

Productivity evidence supports complementarity in specific settings. Controlled deployments in customer support have shown average gains of 14–15% in issues resolved per hour, with larger improvements for less-experienced agents. Engineer surveys report perceived productivity multiples rising from 1.5x to over 2x within a single year of advancing tools.

Risks Remain, Particularly for Entry-Level and Mid-Skill Roles

None of this implies the transition will be painless. Entry-level white-collar positions in coding, analysis, customer service and administrative support appear most immediately pressured. The World Economic Forum’s projections still anticipate roughly 92 million jobs displaced globally by 2030, even as it forecasts a larger number of new roles created. The distribution of gains and losses will depend heavily on whether firms and policymakers treat AI primarily as a cost-cutting tool or as a means to expand output and create new task categories.

The Deutsche Bank findings offer a cautiously optimistic signal: the workers most exposed to the technology are also the most actively positioning themselves to complement it. Whether this pattern persists if large-scale displacement materializes remains an open question. For now, the data suggest that informed concern is translating into adaptation rather than paralysis.

Policy and Corporate Implications

The evidence points toward two priorities. First, accelerating accessible, high-quality upskilling — particularly programs that combine technical AI literacy with domain expertise and human judgment skills. Second, designing adoption strategies inside firms that explicitly aim to augment rather than merely substitute. Historical precedent from earlier waves of computerization shows that outcomes depend as much on organizational choices and complementary investments as on the technology itself.

The AI Anxiety Paradox documented by Deutsche Bank does not prove that displacement risks are illusory. It does indicate that, at this stage of the technology’s diffusion, the labor market’s dominant response is not retreat but recalibration. That distinction matters for both macroeconomic forecasting and for the design of transition policies.

---

References

Anthropic. (2026). Labor market impacts of AI: A new measure and early evidence. https://www.anthropic.com/research/labor-market-impacts

Autor, D. (2024). Applying AI to rebuild middle class jobs (NBER Working Paper No. 32140). National Bureau of Economic Research. https://www.nber.org/papers/w32140

Brynjolfsson, E., Chandar, B., & Chen, R. (2025). Canaries in the coal mine? Six facts about the recent employment effects of artificial intelligence. Stanford Digital Economy Lab. https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/

Deutsche Bank Research. (2026, July 1). I, robot... you, unemployed: How workers are responding to AI fears. Deutsche Bank Research Institute.

Goldman Sachs. (2025, August). How will AI affect the global workforce? Goldman Sachs Research.

International Monetary Fund. (2024). AI will transform the global economy. Let’s make sure it benefits humanity. https://www.imf.org/en/blogs/articles/2024/01/14/ai-will-transform-the-global-economy-lets-make-sure-it-benefits-humanity

World Economic Forum. (2025). Future of jobs report 2025. https://www.weforum.org/publications/future-of-jobs-report-2025/

(Additional supporting studies and updates from 2025–2026 incorporated for context on realized versus potential effects.)