New federal data shows AI-exposed jobs are shrinking while the rest of the labor market grows.
New federal employment data released today shows that jobs flagged as vulnerable to AI displacement are shrinking even as the broader labor market continues to grow. The numbers give the clearest official look yet at whether the AI job displacement story showing up in headlines is actually visible in the underlying economy, and the answer is more complicated than either side of the debate would like.
The Bureau of Labor Statistics figures cover roughly 9 million jobs across 18 occupations identified as AI exposed. Employment in those roles fell 0.2 percent between May 2024 and May 2025. Over the same period, the broader U.S. labor market grew 0.8 percent. That is a real gap, and it lines up with the intuition that has been building for two years now: software is starting to chip away at certain categories of office work, particularly in support, content, and routine analysis.
The part that tends to get lost in the headlines is what the researchers behind the data actually say. They caution that AI is not cleanly identifiable as the main driver of the slowdown in those occupations. Other factors, including post pandemic hiring corrections, interest rate pressure on white collar budgets, and shifts in offshoring patterns, are tangled up in the same numbers. The data shows a divergence, but it does not yet prove causation, and anyone claiming bots are quietly clearing out office floors is reading more into 0.2 percent than the figures can support.
A separate CNBC analysis released alongside the BLS data adds another wrinkle. Of 23 S&P 500 companies that announced AI linked layoffs, 56 percent saw their stock fall afterward. Nike cut 800 workers citing automation and is down nearly 35 percent. Salesforce replaced support engineers with its Agentforce bots and took a similar hit. Investors are not rewarding headcount cuts the way executives seem to expect them to. The skepticism makes sense if you think about it from a fundamentals perspective: swapping experienced humans for early generation agent software is an unproven trade, and capital markets tend to discount unproven trades.
What is worth watching is whether the 0.2 percent dip deepens as more BLS releases come through 2026, or whether displaced workers quietly show up in adjacent roles that current occupation codes do not capture well. Labor statistics are slow to catch new categories of work, and a lot of what AI is doing right now may be reshaping job content rather than eliminating jobs outright. If that is the case, the truer measure of AI's labor impact will not be employment counts at all but wage growth, task composition, and how quickly displaced workers find their next role.
Originally posted on LinkedIn.