OpenAI: 18% of Jobs Threatened by Automation

OpenAI's Analysis on the Impact of AI on the Labor Market

OpenAI recently published a report titled The AI Jobs Transition Framework, led by economist Alex Martin Richmond, which examines the potential impact of artificial intelligence on the labor market. This document analyzes over 900 occupations, representing a total of 153.7 million jobs, nearly the entirety of the labor market in the United States. The study revolves around three main axes: technical exposure to language models, the degree of human necessity in each profession, and the demand sensitivity in response to cost reductions. These criteria are then correlated with real usage data of ChatGPT.

Categorization of Occupations by Automation Risk

In its report, OpenAI classified the 921 occupations studied into four broad categories based on their exposure to automation:

• 18% of occupations are at high risk of short-term automation. These professions, such as data entry operators, accounting clerks, and certain customer service agents, are characterized by high exposure, low human necessity, and inelastic demand.

• 24% of occupations are expected to primarily transform. Although these jobs are highly exposed to AI, they still require human intervention. The productivity gains enabled by AI could lead to workforce reductions without completely eliminating these roles.

• 12% of occupations could see job growth. Professions such as developers or graphic designers exhibit high demand elasticity, which could generate increased demand due to the cost reductions brought about by AI.

• 46% of occupations appear unlikely to undergo significant changes in the short term.

The Gap Between AI Potential and Actual Use

To validate its analysis, OpenAI used anonymized data on professional usage of ChatGPT, collected in the second half of 2025. The goal was to compare theoretical exposure to AI with its actual use in each occupation.

The results revealed a systematic gap: across all sectors, the actual use of AI is significantly lower than its technical potential. OpenAI refers to this phenomenon as capability overhang. In the most exposed occupations to automation, the actual usage rate reaches only 23.8%, while the potential is estimated at 90%, resulting in a gap of 66 points. This disparity highlights the persistent difference between AI capabilities and their concrete adoption.

This gap complicates the assessment of AI's real impact on employment, even when the technical capabilities are already available.