Arul Murugan, Tomás Aguirre, Abhishek Nagaraj, Rishi Bommasani | Preprint | 2026

Website | Paper | Thread | Data | Code

Frontier AI is developed by a small number of US and Chinese companies, but its labor-market exposure is global and uneven. This paper asks how much of each country’s current workforce performs tasks that frontier AI can currently help perform, and whether conclusions drawn from US or European labor markets generalize to the rest of the world.

We combine occupation-level frontier-AI exposure estimates with internationally comparable employment data for 141 countries. The resulting national exposure measure is a composition-weighted view of how strongly a country’s current labor allocation overlaps with work that frontier AI can accelerate or transform. It is not a forecast of wages, adoption, or net job losses; it is a comparable measure of exposure through existing labor-market structure.

Findings

Exposure varies sharply across countries. National labor composition produces large differences in frontier-AI exposure. Countries with more white-collar and service work are generally more exposed, while economies with larger agricultural or manual-work shares tend to be less exposed.

Women are more exposed in most countries. In 91% of the countries we study, women are more exposed than men because of occupational composition, especially concentration in white-collar, clerical, and sales work. The exceptions are countries where women’s employment remains concentrated in agriculture and household-enterprise work.

National exposure predicts AI adoption. The exposure measure predicts country-level adoption statistics from Anthropic, OpenAI, and Microsoft, despite the three sources measuring different forms of AI usage.

Remittances create indirect exposure. Some countries depend heavily on income earned by migrants abroad. When those income flows come from highly exposed destination economies, a country’s effective exposure can rise even if its domestic workforce is less exposed.

Why This Matters

Policy responses calibrated to the United States or Europe will not generalize cleanly across countries. The paper points toward national AI policy that accounts for workforce composition, gendered labor allocation, adoption capacity, and cross-border income dependencies.