Labor Market Impacts

Labor market impacts of AI: A new measure and early evidence \ Anthropic Economic Research Labor market impacts of AI: A new measure and early evidence Mar 5, 2026 Read in PDF

Key findings We introduce a new measure of AI displacement risk, observed exposure , that combines theoretical LLM capability and real-world usage data, weighting automated (rather than augmentative) and work-related uses more heavily AI is far from reaching its theoretical capability: actual coverage remains a fraction of what's feasible Occupations with higher observed exposure are projected by the BLS to grow less through 2034 Workers in the most exposed professions are more likely to be older, female, more educated, and higher-paid We find no systematic increase in unemployment for highly exposed workers since late 2022, though we find suggestive evidence that hiring of younger workers has slowed in exposed occupations

Introduction The rapid diffusion of AI is generating a wave of research measuring and forecasting its impacts on labor markets. But the track record of past approaches gives reason for humility. For example, a prominent attempt to measure job offshorability identified roughly a quarter of US jobs as vulnerable, but a decade on, most of those jobs maintained healthy employment growth. The government’s own occupational growth forecasts, while directionally correct, have added little predictive value beyond linear extrapolation of past trends. Even in hindsight, the impact of major economic disruptions on the labor market is often unclear. Studies on the employment effects of industrial robots reach opposing conclusions, and the scale of job losses attributed to the China trade shock continues to be debated. 1 In this paper, we present a new framework for understanding AI’s labor market impacts, and test it against early data, finding limited evidence that AI has affected employment to date. Our goal is to establish an approach for measuring how AI is affecting employment, and to revisit these analyses periodically. This approach won't capture every channel through which AI could reshape the labor market, but by laying this groundwork now, before meaningful effects have emerged, we hope future findings will more reliably identify economic disruption than post-hoc analyses. It is possible that the impacts of AI will be unmistakable. This framework is most useful when the effects are ambiguous—and could help identify the most vulnerable jobs before displacement is visible. Counterfactuals Causal inference is easier when the effects are large and sudden. The COVID-19 pandemic and accompanying policy measures caused economic disruption so stark that sophisticated statistical approaches were unnecessary for many questions. For example, unemployment jumped sharply in the early weeks of the pandemic, leaving little room for alternative explanations. The impacts of AI, however, might be less like COVID and more like the internet or trade with China. The effects may not be immediately clear from aggregate unemployment data; factors like trade policy and the business cycle could cloud interpretations of trend lines. One common approach is to compare outcomes between more or less AI-exposed workers, firms, or industries, in order to isolate the effect of AI from confounding forces. 2 Exposure is typically defined at the task level: AI can grade homework but not manage a classroom, for example, so teachers are considered less exposed than workers whose entire job can be performed remotely. Our work follows this task-based approach, incorporating measures of theoretical AI capability and real-world usage, before aggregating to occupations. 3 Measuring exposure Our approach combines data from three sources. The O*NET database , which enumerates tasks associated with around 800 unique occupations in the US. Our own usage data (as measured in the Anthropic Economic Index ). Task-level exposure estimates from Eloundou et al. (2023), which measure whether it is theoretically possible for an LLM to make a task at least twice as fast.

Eloundou et al.’s metric, β, scores tasks on a simple scale: 1 if a task can be doubled in speed by an LLM alone, 0.5 if it requires additional tools or software built on top of the LLM, and 0 otherwise. 4 Why might actual usage fall short of theoretical capability? Some tasks that are theoretically possible may not show up in usage because of model limitations. Others may be slow to diffuse due to legal constraints, specific software requirements, human verification steps, or other hurdles. For example, Eloundou et al. mark “Authorize drug refills and provide prescription information to pharmacies” as fully exposed (β=1). We have not observed Claude performing this task, although the assessment seems correct in that it could theoretically be sped up by an LLM. That said, these measures of theoretical capability and actual usage are highly correlated. As Figure 1 shows, 97% of the tasks observed across the previous four Economic Index reports fall into categories rated as theoretically feasible by Eloundou et al. (β=0.5 or β=1.0). Figure 1: Share of Claude usage by Eloundou et al. task exposure rating This figure shows Claude usage distributed across O*NET tasks grouped by their theoretical AI exposure. Tasks rated β=1 (fully feasible for an LLM alone) account for 68% of observed Claude usage, while tasks rated β=0 (not feasible) account for just 3%. Data on Claude usage comes from the previous four Economic Index reports. A new measure of occupational exposure Our new measure, observed exposure , is meant to quantify: of those tasks that LLMs could theoretically speed up, which are actually seeing automated usage in professional settings? Theoretical capability encompasses a much broader range of tasks. By tracking how that gap narrows, observed exposure provides insight into economic changes as they emerge. Our measure qualitatively captures several aspects of AI usage that we think are predictive of job impacts. A job's exposure is higher if: Its tasks are theoretically possible with AI Its tasks see significant usage in the Anthropic Economic Index 5 Its tasks are performed in work-related contexts It has a relatively higher share of automated use patterns or API implementation Its AI-impacted tasks make up a larger share of the overall role 6

We give mathematical details in the Appendix . We count tasks that are theoretically capable with an LLM as…