NBER: AI-written dissertations tied to leaving academia

AI-generated text was absent from US STEM dissertation abstracts before 2023 and appeared in 29.4% of abstracts filed during the first five months of 2026, according to a National Bureau of Economic Research working paper covered by Nature. The paper, by Daniel P. Gross, Dror Shvadron and Hansen Zhang, ran the detection service Pangram over more than 226,000 abstracts filed between January 2019 and May 2026.

The share rose from 1.7% in 2023 to 18.8% in 2025 before reaching 29.4% for early 2026. Mathematical and computer sciences led fields at 36%. Graduates educated in non-English-speaking countries filed AI-written abstracts at 41%, against 23% for those from English-speaking countries.

The career finding is the one that is travelling. Among US-origin PhD graduates in the same program and cohort, those whose dissertations contain AI writing were 17% less likely to be employed in academia and 16% more likely to be employed in industry. The association was less pronounced for international students.

"If our students are only doing half the work and they're dishing the rest to AI, do they really have the depth of understanding" to be good scientists, Gross, an NBER economist based in Durham, asks in the Nature piece. The working paper itself puts the concern more flatly: AI "appears to also be displacing a potentially important portion of the work through which scientific expertise is developed."

Ang Liu, an engineer at the University of New South Wales, told Nature that "rigorous research training becomes even more important" when research moves into areas where the answers are not already known.

The study measures AI use in abstracts only, and the headline correlation is within-program and within-cohort rather than causal: it does not establish that AI writing drove the career divergence rather than tracking students who already preferred industry.