New data from the US Bureau of Labor Statistics shows that 18 occupations classified as “highly exposed” to AI account for roughly 10 million jobs. From May 2024 to May 2025, employment across that group slipped 0.2%, while total US employment grew 0.8% over the same period. The gap is now visible in the numbers.
Bloomberg’s breakdown adds an important qualifier. If “medical secretaries and administrative assistants” are excluded because hiring in healthcare kept rising, the other 17 occupations posted a 1.6% decline from 2024 to 2025. That marked the second consecutive year with the same rate of contraction, pointing to an ongoing shift rather than a one-off drop.
Customer service, administrative support and sales were hit hardest
The steepest one-year decline came from customer service representatives, which lost 130,180 jobs, a 4.8% decrease. Secretaries and administrative assistants, excluding medical, legal and executive roles, fell by 31,030, or 1.8%. Wholesale and manufacturing sales representatives, excluding technical and scientific products, declined by 28,670, down 2.3%.
The longer view is sharper. Since May 2022, the last full data point before ChatGPT arrived, credit authorizers, auditors and clerks have fallen 26.2%; radio announcers and DJs are down 20.8%; and sales engineers have dropped 13.2%. Many of these roles share the same profile: predictable inputs, standardized outputs and workflows that can be documented with little ambiguity.
Where AI replaces tasks and where it only speeds them up
The report draws a line between jobs that AI can replace and jobs that AI mainly assists. In the first group, AI can handle core tasks directly: taking customer calls, processing refund requests, answering routine questions, reviewing credit files and entering form data. Those duties are easier to automate. They depend less on judgment and more on repetition.
That is different from occupations where AI acts as a tool rather than a substitute. Doctors using AI to read pathology slides, lawyers sorting case precedents with AI, and engineers generating test code are examples cited in the report. In those cases, the main decision still sits with a human worker, even if the process becomes faster.
Entry-level jobs are narrowing first
Bloomberg notes that many of the affected roles have long served as entry points for younger workers. Customer support, data entry, billing, basic financial analysis and junior software support have traditionally been where recent graduates built early experience. This time, those entry-level openings are the ones tightening first.
A Goldman Sachs report from April 2026 estimated that AI was causing a net reduction of about 16,000 US jobs per month. It also said AI replaced about 25,000 positions every four weeks, while AI-augmented roles added back roughly 9,000. The pattern described by the data is not a sudden collapse. It is a slow structural substitution unfolding month by month, with customer service centers, content moderation, legal support administration, billing functions and early-stage software support among the areas under the most pressure.