NBER: nearly 30% of 2026 US STEM PhD abstracts show AI writing

Nearly three in ten US STEM PhD dissertations filed in the first five months of 2026 contain AI-generated text in their abstracts, Nature reported on a new NBER working paper. The same graduates were 17% less likely than their peers to pursue academic careers, and 16% more likely to work in industry.

The study, by Durham-based economist Daniel Gross and colleagues, ran the AI-detection service Pangram across more than 226,000 dissertations completed at US universities between January 2019 and May 2026. AI text first surfaced in abstracts in 2023 at 1.7%, climbed to 18.8% in 2025, and reached 29.4% in the first five months of 2026. In mathematical and computer sciences the rate was 36%. Roughly 90% of the flagged abstracts were classified as entirely AI-generated.

The language split is sharp: 41% of dissertations from non-English-speaking countries contained AI text, against 23% from English-speaking countries.

Ang Liu, an engineer at the University of New South Wales, told Nature that AI assistance becomes problematic when research enters areas with incomplete knowledge. "That is where rigorous research training becomes even more important," he said.

The paper does not establish whether AI use causes the shift away from academic careers or simply tracks students who were already headed for industry. Pangram's false-positive rate on scientific prose, especially from non-native English writers, is also not spelled out in the reporting.