The first hard warning is early-career hiring. It is not yet a clean AI verdict.

What the study measured

Lee C. Tucker's Census working paper uses quarterly administrative tabulations for private employers in 45 states, through 2025 Q2. It maps occupational AI exposure into detailed industry-state cells and compares early-career workers with lower-exposure cells around ChatGPT's November 2022 release.

The result

Early-career hires in the most-exposed cells fell immediately by 9% relative to less-exposed cells. Regression-adjusted employment in the highest-exposure quintile was 12% lower after 10 quarters. The paper says the hiring reduction explains the subsequent employment decline and finds no catch-up hiring that returns those cells to their earlier employment level.

Why this is not proof that AI did it

The study does not observe a specific employer adopting an AI system and compare it with a matched employer that did not. It uses an exposure score. The author finds earlier shifts around the COVID period and identifies remote work and educational attainment as correlated alternative explanations. The discontinuity in hiring at ChatGPT's release is difficult to dismiss, but it does not remove those alternatives.

Cross-checks that narrow the story

A separate Census firm survey shows adoption is uneven and most AI-using firms report augmentation only; 2% reported an AI-related employment decrease in its reference period. In Denmark, Humlum and Vestergaard find task reorganization but no detectable earnings or recorded-hours effect larger than 2% over the first two years. Neither result answers the U.S. early-career hiring question. They show why a single “AI is taking jobs” line is too crude.

What would change this conclusion

• Replication with observed firm-level AI adoption and stronger controls for remote work, education and other post-pandemic changes.
• Updated data showing whether entry-level hiring recovers, worsens or spreads to other workers and sectors.
• Evidence that separates task reorganization from changes in wages, hours, job quality and employment.