MIT report: AI can already replace nearly 12% of the U.S. workforce
Artificial intelligence is now advanced and cheap enough to perform work equal to nearly 12% of U.S. jobs, according to a new MIT study—news that’s likely further to intensify pressure on employers, workers, and policymakers to prepare for rapid shifts in business and the economy.
MIT’s research, written in October but released on Wednesday, estimates that current AI systems could already take over tasks tied to 11.7% of the U.S. labor market, representing about 151 million workers and roughly 11.7% of total wage value, or around $1.2 trillion in pay. Unlike earlier estimates that focused on theoretical “exposure” to automation, the MIT research focuses on jobs where AI can perform the same tasks at a cost that’s either competitive with or cheaper than human labor.
The findings come from Project Iceberg, a large-scale labor simulation developed by MIT in collaboration with Oak Ridge National Laboratory, home to the Frontier supercomputer.
The model creates what researchers describe as 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 across 923 job types in 3,000 counties and maps them against what current AI systems can already do.
“We’re effectively creating a digital twin of the U.S. labor market,” Prasanna Balaprakash, a director at Oak Ridge National Laboratory and co-leader of the study Tod NBC.
A KEY CAVEAT
MIT’s report makes it clear that the 11.7% figure reflects technical capability and economic feasibility, not a prediction that those jobs will disappear on a set timetable. It also highlights a gap between what is visible today and what is possible.
AI adoption so far has been concentrated in tech work, particularly coding, representing about 2.2% of wage value, or roughly $211 billion in pay. But the researchers find that AI is already capable of handling cognitive and administrative tasks across finance, healthcare, and professional services that together represent around $1.2 trillion in wages—about five times the currently visible impact.
Early analysis points to significant exposure in white-collar, knowledge-heavy fields that were once seen as relatively insulated 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 execute many routine tasks. In other words, much of the potential disruption sits in more traditional back-office and professional roles that have drawn less public attention in AI debates.
At the same time, MIT researchers and other economists caution that capability does not automatically translate into widespread job losses. Earlier 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 near term, even where the technology could perform the tasks. Separate research from MIT Sloan concluded that, from 2010 to 2023, AI exposure did not lead to broad net job losses and often coincided with faster revenue and employment growth at adopting firms.
The Iceberg Index is not designed to forecast specific layoffs. Instead, it gives policymakers and business leaders a way to stress-test different scenarios before they commit training dollars, infrastructure spending or new regulations. Tennessee, North Carolina and Utah have already begun using the platform to evaluate how AI might reshape their workforces and to inform state-level AI workforce action plans, the MIT report said.
For companies, the study illustrates that the window to treat AI as a distant future issue is closing. For governments, it raises practical questions about how to retrain workers, support regions and sectors with high exposure, and adapt tax and social safety net systems to a labor market where software can already do a meaningful share of the work.
The findings come from Project Iceberg, a large-scale labor simulation developed by MIT in collaboration with Oak Ridge National Laboratory, home to the Frontier supercomputer.
The model creates what researchers describe as 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 across 923 job types in 3,000 counties and maps them against what current AI systems can already do.
“We’re effectively creating a digital twin of the U.S. labor market,” Prasanna Balaprakash, a director at Oak Ridge National Laboratory and co-leader of the study Tod NBC.
A KEY CAVEAT
MIT’s report makes it clear that the 11.7% figure reflects technical capability and economic feasibility, not a prediction that those jobs will disappear on a set timetable. It also highlights a gap between what is visible today and what is possible.
AI adoption so far has been concentrated in tech work, particularly coding, representing about 2.2% of wage value, or roughly $211 billion in pay. But the researchers find that AI is already capable of handling cognitive and administrative tasks across finance, healthcare, and professional services that together represent around $1.2 trillion in wages—about five times the currently visible impact.
Early analysis points to significant exposure in white-collar, knowledge-heavy fields that were once seen as relatively insulated 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 execute many routine tasks. In other words, much of the potential disruption sits in more traditional back-office and professional roles that have drawn less public attention in AI debates.
At the same time, MIT researchers and other economists caution that capability does not automatically translate into widespread job losses. Earlier 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 near term, even where the technology could perform the tasks. Separate research from MIT Sloan concluded that, from 2010 to 2023, AI exposure did not lead to broad net job losses and often coincided with faster revenue and employment growth at adopting firms.
The Iceberg Index is not designed to forecast specific layoffs. Instead, it gives policymakers and business leaders a way to stress-test different scenarios before they commit training dollars, infrastructure spending or new regulations. Tennessee, North Carolina and Utah have already begun using the platform to evaluate how AI might reshape their workforces and to inform state-level AI workforce action plans, the MIT report said.
For companies, the study illustrates that the window to treat AI as a distant future issue is closing. For governments, it raises practical questions about how to retrain workers, support regions and sectors with high exposure, and adapt tax and social safety net systems to a labor market where software can already do a meaningful share of the work.