NBER: AI-written STEM PhD abstracts tied to industry careers

Roughly 30% of US STEM doctoral dissertation abstracts filed in the first five months of 2026 contained AI-generated text, up from 1.7% in 2023, and the graduates whose abstracts triggered the detector were 17% less likely to be working in academia and 16% more likely to be in industry, according to a working paper covered by Nature. The analysis, co-authored by Daniel Gross, an economist at NBER based in Durham, ran the Pangram AI-detection service across more than 226,000 STEM doctoral graduates who completed their degrees between January 2019 and May 2026 at US universities.

Mathematical and computer sciences led, with 36% of abstracts flagged. The language gap was wider still: around 41% of dissertations by graduates educated in non-English-speaking countries carried AI-generated abstracts, compared with 23% from English-speaking countries. The 17/16-point academia-versus-industry gap applied specifically to graduates who had done their undergraduate degrees in the US.

The split is a correlation, not a demonstrated cause, and the paper does not separate wholly from partially AI-generated writing, though Gross says 90% of the flagged abstracts were classified by Pangram as entirely AI-generated. He framed the concern in training terms. "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, he asked. Ang Liu, an engineer at the University of New South Wales in Sydney, Australia, told Nature that "rigorous research training becomes even more important" when researchers are exploring areas where answers are not already known.

Two researchers we follow in our Who's Who directory shared the paper.