Artificial intelligence is now advanced and inexpensive enough to do work that accounts for nearly 12% of U.S. jobs, according to a new MIT study – news that is likely to add even more pressure on employers, workers and policymakers to prepare for rapid changes in business and the economy.
The MIT study, written in October but released Wednesday, estimates that current AI systems could already handle tasks associated with 11.7% of the U.S. labor market, representing about 151 million workers and about 11.7% of total wage value, or about $1.2 trillion in wages. Unlike previous estimates that focused on theoretical “exposure” to automation, MIT research focuses on jobs where AI can perform the same tasks at a price that is either competitive with human labor or cheaper.
The results come from Project Iceberg, a large-scale work simulation developed by MIT in collaboration with Oak Ridge National Laboratory, home of the Frontier supercomputer.
The model creates what researchers call a “digital twin of the U.S. labor market,” simulating 151 million workers as individual agents, each with specific skills, occupations and locations. It tracks more than 32,000 skills in 923 job types in 3,000 counties and maps them to what current AI systems can already do.
“We are effectively creating a digital twin of the U.S. labor market,” said Prasanna Balaprakash, director at Oak Ridge National Laboratory and co-leader of the study CNBC.
An important caveat
The MIT report makes clear that the 11.7% figure reflects technical capability and economic feasibility, not a prediction that these jobs will disappear on a set schedule. It also highlights the gap between what is visible today and what is possible.
To date, AI adoption has been focused on technical work, particularly programming, and accounted for about 2.2% of wage value, or about $211 billion in wages. However, the researchers note that AI is already capable of handling cognitive and administrative tasks in finance, healthcare and professional services, which together account for around $1.2 trillion in wages – about five times the impact currently seen.
Initial analyzes indicate significant strain in knowledge-intensive white-collar sectors that were once considered relatively isolated from automation. Finance, healthcare administration, human resources, logistics, and professional services such as legal and accounting work are among the areas where existing AI tools, including large language models (LLMs) and other software agents, can already perform many routine tasks. In other words, much of the potential disruption lies in more traditional back-office and professional roles, which have received less public attention in the AI debate.
At the same time, MIT researchers and other economists warn that performance does not automatically lead to widespread job losses. Previous work from MIT's Computer Science and Artificial Intelligence Laboratory found that for many roles, fully replacing human workers with AI remained too expensive or impractical in the short term, even if the technology could accomplish the tasks. A separate study from MIT Sloan concluded that AI exposure from 2010 to 2023 did not result in overall net job losses and was often accompanied by faster sales and employment growth at acquiring companies.
The Iceberg Index is not designed to predict specific layoffs. Instead, it offers policymakers and business leaders the opportunity to stress test various scenarios before committing to training dollars, infrastructure spending or new regulations. Tennessee, North Carolina and Utah have already begun using the platform to assess how AI could transform their workforces and create AI workforce action plans at the state level, the MIT report said.
For companies, the study makes it clear that the window of opportunity to treat AI as a topic of the distant future is closing. For governments, it raises practical questions about how to retrain workers, support high-risk regions and sectors, and adapt tax and social security systems to a labor market where software can already do a meaningful share of the work.
For this story, Assets used generative AI to help with an initial draft. An editor checked the accuracy of the information before publication.