For three years, the conventional wisdom on AI and wages ran roughly like this: automation would hit lower-paid, lower-skilled workers first. Truck drivers. Cashiers. Data entry clerks. Degree holders in knowledge work would be fine -- maybe even advantaged, as AI became a tool they'd use rather than a threat they'd face.
New data released this month dismantles that assumption.
A working paper from the U.S. Census Bureau, published in September 2026, tracked actual labor market outcomes for recent college graduates across every major since large language models became widely available in late 2022. The findings are blunt. The most AI-exposed college majors -- computer science, accounting, journalism, and engineering -- saw their likelihood of landing an initial job fall by five percentage points and their full-quarter starting earnings fall by thirteen percent.
Those are not projected losses. Those are already in the data.
Which Graduates Are Getting Hit
The Census researchers divided college majors into deciles by AI exposure level. The outcomes diverge sharply at the top. Graduates in the most AI-exposed decile have seen their early career trajectories bend downward in two ways simultaneously: they're less likely to find employment at all right out of school, and when they do find work, they're earning meaningfully less.
About half of the earnings decline came from lower pay within the same industries. The other half came from graduates moving into lower-paid sectors entirely -- restaurants, retail, service jobs that did not require their degrees. That second dynamic is worth paying attention to. It is not just that AI-exposed graduates are getting smaller paychecks in tech or finance. Some of them are leaving those fields altogether.
The majors that came out as least AI-exposed? Nursing and education. The workers in persistent high demand -- bedside nurses, special education teachers, physical therapists -- turn out to have natural insulation from the AI wave because their work is fundamentally physical, relational, or requires real-time human judgment in unpredictable settings.
Economists Are Shifting Their Expectations
If the Census paper were an isolated finding, it could be explained away as a temporary market dislocation. But a separate survey released this week suggests that professional economists now view this wage pressure as durable and growing.
Indeed Hiring Lab conducts a quarterly survey of professional economists on the labor market, administered in partnership with Pulsenomics. The Q3 2026 edition fielded September 8-16 and drew responses from 123 panelists.
The survey tracks a diffusion index measuring whether economists expect AI to weigh more heavily on the wages of college-educated or non-college-educated workers. A reading above 50 means economists expect more pressure on college grads. In Q2, the index sat at 47.3 -- essentially neutral. In Q3, it dropped to 42.4.
That is the most pronounced quarter-over-quarter shift in the survey's history. More economists, not fewer, now expect AI's wage bite to land harder on degree holders. The panel sees essentially no meaningful AI wage pressure in either direction for non-college-educated workers.
The story being told by two separate data sources -- Census administrative records and a 123-economist survey -- is the same story.
The Seniority Trap in the Advertised Pay Numbers
At this point you might be wondering: what about those reports showing that AI-exposed job postings are paying significantly more? Those are real, but they require a closer read.
Indeed Hiring Lab's own research published on September 17 found that advertised pay in the most AI-exposed occupations is up roughly 46% since 2021, compared to 25% for less-exposed occupations. That sounds good for degree holders in AI-adjacent fields. Here is what the same research also found:
At the senior level, AI-exposed pay grew 45% from its 2021 baseline through mid-2026. At the mid-level, there is a clear but smaller gap. At the entry level, there is a 2-point gap.
The premium exists. It is almost entirely concentrated at senior workers who already have demonstrable AI skills. For new graduates entering AI-adjacent fields without a track record of applied AI work, the premium is effectively nonexistent. And the Census data suggests that for that same population, outcomes have deteriorated since ChatGPT launched.
What the advertised salary data cannot show you is what happens to graduates who don't get the AI-adjacent roles at all -- who end up in the service sector with a CS degree because the entry-level tech market absorbed fewer of them than it used to. That compression is invisible in job posting data and visible in administrative earnings records.
What This Means for Your Comp Benchmarks
Compensation surveys are lagged by design. The data they reflect was collected months ago, and it is skewed toward workers who are already employed in their fields. It does not fully capture the shift happening at the entry-level intake for AI-exposed majors.
If you are hiring entry-level roles in software development, data and analytics, accounting, marketing, or finance -- fields that rank high on AI-exposure indexes -- your comp benchmarks may be running above where the actual market is clearing today for new grads. That does not mean to low-ball candidates. It means to verify your benchmarks against recent hires, not only published surveys.
For senior AI-skilled roles, the opposite is true. That market is genuinely competitive. The 46% pay growth in advertised senior AI-exposed postings reflects real scarcity. If you are recruiting experienced engineers, data scientists, or finance professionals who can document applied AI fluency, expect the market to push back hard on anything below the going rate.
A practical framework:
- Entry-level, AI-exposed field + no documented AI experience: verify your comp floor against what recent grads are actually accepting, not what a 2025 survey reported
- Entry-level, AI-exposed field + demonstrable AI skills: treat them like a mid-level hire; the market data supports a premium
- Senior, AI-skilled, proven: this is the true premium tier -- don't anchor to pre-2024 benchmarks
The Degree Is No Longer a Proxy
The older model of compensation benchmarking treated the college degree as a reliable floor signal. A CS degree meant you were a qualified technical candidate who warranted tech-sector pay. An accounting degree meant you were entering a credentialed profession with a defined pay band.
That model worked when a degree reliably conferred the skills needed to do the job. AI has changed what "skills needed to do the job" means in those exact fields faster than any curriculum can track. A 2025 accounting grad who cannot use AI for audit workflows is less competitive than a 2020 accounting grad who spent five years using them. A journalism grad who cannot produce AI-augmented content faster and better than the tool alone is competing against the tool itself.
The workers who are insulated -- nurses, tradespeople, physical therapists -- are insulated precisely because their skills cannot be verified on a resume. They have to be demonstrated in person, in real time, under conditions that vary unpredictably. No language model can apprentice through that.
For recruiters, the practical implication is simple: stop using the college degree as a proxy for competency in AI-exposed roles. Test for it. Build structured assessments that require candidates to demonstrate applied AI use, not just list it. The degree tells you less about capability than it did three years ago, and the comp benchmarks built on degree-based assumptions are drifting away from market reality.
If you recruit in technology, finance, or professional services and want a faster signal on where the comp floor is actually clearing today, BlueLine's tools can help you benchmark against live market data instead of lagged surveys.