PhDs with AI-written dissertations less likely to stay academic

Nearly 30% of US STEM PhD dissertation abstracts filed in the first five months of 2026 contain AI-generated writing, according to a working paper reported in Nature. The share was 1.7% in 2023, 18.8% in 2025, and 29.4% in early 2026.

The paper, NBER Working Paper 35859, is by Daniel P. Gross and Hansen Zhang of Duke University and Dror Shvadron of the University of Toronto. The team examined more than 226,000 STEM dissertations completed at US universities between January 2019 and May 2026, flagging abstracts via the Pangram AI-detection service at a threshold of at least 2% AI-written text.

Math and computer science led the fields at 36% flagged. The split by educational background was wider still: among graduates educated in non-English-speaking countries, 41% of dissertations contained AI-generated abstracts, compared with 23% for those from English-speaking countries.

Among US-educated PhD recipients, those whose abstracts were flagged for AI writing were 17% less likely to work in academia and 16% more likely to work in industry than peers who did not use AI. The authors argue that "delegating research writing to AI may undermine the development of expertise that scientific training is supposed to build."

They do not claim causation: students already headed for industry may simply put less effort into prose craftsmanship. The paper has not been peer-reviewed. Two researchers we follow were already circulating the Nature write-up the day it posted.