AI labor disruption remains modest but is harder to ignore at the occupational level

Overview

This report updates Morgan Stanley's AI labor disruption tracker, covering data through the first half of 2026. The core conclusion is that AI's impact on the aggregate labor market remains small, resembling a phenomenon that is "strong at the micro level but weak at the macro level"; however, signals of occupational-level unemployment, reemployment difficulty, and task changes are becoming clearer.

Core views

The report finds that AI-related labor disruption has not yet constituted a large-scale macroeconomic employment shock, but unemployment in occupations with high AI exposure continues to rise. After controlling for differences in occupational cyclical sensitivity, the deviation from normal levels has reached approximately 0.5 percentage points and is statistically significant at the 90% confidence level. High-exposure occupations account for approximately 30% of employment, implying an upper-bound contribution of approximately 15bp to the overall unemployment rate. By contrast, industry-level employment, job openings, and corporate commentary still do not support a strong narrative of broad-based substitution.

Analysis framework

The report uses an AI disruption tracker that combines labor-market microdata, worker flows, task reallocation, wage and employment data, job openings, and analysis of corporate earnings-call transcripts, comparing occupational and industry performance by AI exposure. The main results use the AI exposure measure of Felten et al. (2021) and adjust for the historical cyclicality of different occupations and industries to identify abnormal deviations relative to aggregate labor-market conditions.

Methodology notes

• Labor market trackingAI disruption tracker

  AI exposure grouping and labor-market indicator panel

  Occupations and industries are grouped by AI exposure to track unemployment, worker flows, task changes, wage and employment data, job openings, and corporate text, in order to infer AI's potential impact on the labor market.

• Econometric adjustmentCyclically adjusted residual analysis

  Residuals from regressions on the aggregate unemployment rate or aggregate employment trend

  Because occupations and industries differ in their inherent cyclical sensitivity, the report uses overall labor-market conditions to explain unemployment or employment performance in each group, then examines whether the residuals are abnormal to reduce bias from cyclical factors.

• AI exposure measurementFelten et al. (2021) AI exposure measure

  Occupational AI exposure

  The report's main results use the AI exposure indicator developed by Felten et al. to measure the potential extent to which different occupations are affected by AI, and notes that an alternative measure from Morgan Stanley's Thematic team is used for robustness checks in the appendix.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• U.S. labor market

  Rising unemployment in occupations with high AI exposure may exert modest upward pressure on the overall unemployment rate.

  Strengths

  The aggregate impact remains small, and there has been no broad-based employment collapse at the industry level.

  Weaknesses

  Reemployment has become more difficult in occupations with high AI exposure, and unemployment durations have lengthened.

  Comparison

  Signals are stronger at the occupational level than at the industry level; signals among younger workers are stronger than those among workers overall.

  Risks

  If AI adoption accelerates, occupational-level disruption could transmit further into the aggregate unemployment rate.

• AI-related industries and the technology diffusion theme

  AI is affecting corporate labor structures through task reorganization and productivity improvements.

  Strengths

  Corporate commentary places greater emphasis on productivity and process efficiency, indicating continued positive efficiency potential from AI applications.

  Weaknesses

  Some jobs may face substitution or slower hiring in the near term.

  Comparison

  At present, corporate text and industry employment data provide weaker support for "broad-based layoffs" than micro-level occupational data.

  Risks

  If companies shift from efficiency improvements toward explicit workforce reductions, the market's assessment of AI's macroeconomic externalities could change.

Key data

• Estimated impact of AI on the overall unemployment rateAt most approximately 15bpAs of June 2026, this is a rough upper bound calculated by multiplying the abnormal unemployment rate in high-AI-exposure occupations by their approximately 30% employment share.
• Cyclically adjusted unemployment deviation in high-AI-exposure occupationsApproximately 0.5 percentage pointsUp from approximately 0.3 percentage points in the previous report, with the signal more pronounced in 1H26.
• Employment share of high-AI-exposure occupationsApproximately 30%Used to estimate the upper-bound contribution to the overall unemployment rate.
• Statistical significance90% confidence levelThe unemployment residual for high-AI-exposure occupations has exceeded the two-standard-error range; occupations with medium exposure remain statistically insignificant.
• Key affected age groupAges 22-27Younger workers show the most pronounced increase in unemployment in high-AI-exposure occupations.

Impact & implications

The report suggests that AI is more clearly changing employment matching and work tasks in certain occupations, but has not yet generated a significant employment shock in aggregate macroeconomic data. For macro research, this means that near-term upward pressure on the unemployment rate remains limited; for industry and company research, attention should focus on hiring, layoffs, task reorganization, and productivity gains in high-AI-exposure jobs rather than relying solely on aggregate industry employment to assess AI substitution.

Risks

• Omitted-variable risk: monetary policy, tariffs, immigration policy, and post-pandemic hiring normalization may also affect the performance of different occupations.
• Micro-level occupational signals do not yet fully prove that AI is the cause; some differences may result from adjustments specific to the occupations or industries themselves.
• The sample size of younger non-college-educated workers is small, resulting in a lower signal-to-noise ratio and requiring cautious interpretation.
• Industry-level and job-opening data remain noisy, which may obscure more granular AI effects or indicate that the effects are not yet significant.

What to watch

• Whether unemployment in occupations with high AI exposure continues to rise relative to occupations with low and medium exposure.
• Whether the probability of transitioning from unemployment to employment and the layoff rate in high-AI-exposure occupations continue to deteriorate or return to normal.
• Whether the unemployment gap among workers aged 22-27 across different education groups widens.
• Whether task-change indicators reaccelerate after a temporary decline in 1H26.
• Whether detrended employment in industries with high AI exposure weakens further.
• Whether the proportion of corporate earnings-call commentary linking AI with layoffs, hiring freezes, and productivity improvements changes.