Yale Budget Lab: The Labor Market Data Shows AI Still Isn't Disrupting Jobs at Scale

By Hector Herrera | May 2, 2026 | Work

Three and a half years after ChatGPT launched, Yale's Budget Lab analyzed the most current U.S. labor market data and found no measurable economy-wide employment disruption attributable to AI. The findings complicate the dominant narrative — and raise pointed questions about whether companies citing AI for mass layoffs are telling the truth.

The Yale Budget Lab's March 2026 analysis used the Current Population Survey (CPS) — the same dataset the Bureau of Labor Statistics relies on for official unemployment figures — to track occupational dissimilarity and AI exposure metrics since ChatGPT's November 2022 launch. Both measures remain essentially flat. If AI were restructuring the labor market at the scale suggested by corporate announcements and media coverage, these indicators would show it. They don't.

What the Data Actually Shows

The Budget Lab's methodology tracks two key signals:

Occupational dissimilarity — the degree to which workers are shifting from one type of job to another. If AI were displacing workers from knowledge work into different roles, this measure would climb. It hasn't moved meaningfully.

AI exposure metrics — the share of work tasks within each occupation that AI tools could plausibly perform. These scores, built from prior research frameworks, have remained stable even as AI capabilities have improved significantly. The jobs most exposed to AI — paralegal work, data analysis, customer service, content drafting — continue to employ roughly the same number of people as before.

The report also draws on Anthropic usage data that shows AI interactions are more likely associated with task automation (handling a specific step in a workflow) than worker replacement (eliminating a position entirely). People are using AI to do parts of their jobs faster. They're mostly keeping their jobs.

The 'AI-Washing' Question

The findings sharpen a question that has circulated in labor economics circles but rarely receives this kind of empirical grounding: Are companies using "AI" as cover for cost-cutting decisions that have other causes?

Between 2024 and 2026, dozens of large employers — including major banks, tech firms, and media companies — announced layoffs citing AI-driven efficiency gains as a primary rationale. Those announcements typically sent stock prices higher, as investors interpreted them as evidence of disciplined cost management tied to a tech upgrade.

Yale's data suggests a different interpretation is at least plausible. If economy-wide labor disruption isn't showing up in the data, then the companies announcing AI-driven workforce reductions may be restructuring for conventional reasons — slowing revenue, strategic pivots, post-pandemic normalization — while crediting AI because the narrative is more palatable to investors and easier to communicate publicly than "we hired too many people in 2021."

This doesn't mean AI isn't affecting work. What it means is that the effect, at the aggregate level, looks more like productivity enhancement than displacement. Workers are doing more with the same number of hours, not being replaced by software.

Why the Mismatch Exists

Several factors could explain the gap between corporate AI-displacement narratives and the aggregate data:

• Concentration: AI-driven displacement may be real but concentrated in specific job categories or companies — enough to generate headlines but not yet large enough to move economy-wide metrics.
• Lag: Labor market disruptions often take years to show up in survey data. Workers who lose AI-adjacent jobs may be quickly absorbed into other roles, masking the underlying churn.
• Measurement limits: The CPS captures employment status, not the quality or nature of work. Workers doing fundamentally different tasks within the same job title look identical in the data.
• Real displacement ahead: The Budget Lab explicitly notes this is a snapshot, not a projection. The absence of disruption through March 2026 does not mean the same will be true in 2028.

What This Means for Workers and Employers

For workers: The data is reassuring in the short term. If you're in a knowledge-work role, the labor market disruption you've been warned about hasn't materialized at scale. That doesn't mean it won't. It means there is time to adapt — and that adaptation, right now, looks more like learning to use AI tools effectively than preparing for job elimination.

For employers and HR teams: The AI-washing question has real exposure. If a company announces AI-driven workforce reductions, gets a stock bump, and the claimed efficiency gains don't materialize, that gap will eventually be visible in earnings and in litigation. Using AI as a narrative cover for conventional restructuring is a strategy with a shelf life.

For policymakers: The Yale data argues against rushing emergency labor protection legislation based on projected AI displacement that hasn't yet shown up in the data. It also argues for continued monitoring — the picture in 2028 may look very different.

What to Watch

The Budget Lab says it will continue updating this analysis quarterly as new CPS data becomes available. The key signal to track: whether occupational dissimilarity metrics begin to climb in Q3 or Q4 2026, as the latest generation of agentic AI tools — which automate multi-step workflows rather than single tasks — become more widely deployed in enterprise settings. If disruption is coming, that's where it will first appear in the data.

Source: Yale Budget Lab