Daniel P. Gross, Dror Shvadron and Hansen Zhang estimate, in a working paper that has not been peer-reviewed, that AI-generated writing appears in 29% of US STEM PhD dissertations filed in 2026, up from none before 2023. The estimate rests on the commercial Pangram detector, applied mainly to abstracts, and a smaller, unpublished analysis commissioned for Daily Nous reports a rising share in philosophy.
How this is computed
- Key claims supported: 3 of 6
- Type of source: Other
- Independent confirmation: None
Rates the source, not the topic and not our analysis.
Why it matters
Our reading: If the detector holds up, the figures suggest AI-assisted writing has entered doctoral work faster than departmental policies have adapted. The links the researchers report to career paths and student background are correlations, and they could partly reflect detector bias rather than actual use. The debate is therefore as much about measurement and authorship as about adoption.
What the source says
- Conditionally supported The researchers’ working paper reports no AI writing in US STEM dissertations before 2023 and AI writing in 29% of those filed in 2026, based on Pangram scoring of abstracts.
- Conditionally supported According to the same paper, AI writing is more common among students from non-English-speaking countries and at lower-ranked programs. Commenters point to research suggesting detectors may flag edited or translated text from multilingual writers more often.
- Conditionally supported The researchers report that, within the same program and cohort, graduates whose dissertations contain AI writing are less likely to enter academic research careers and more likely to go into non-tenure-track roles or industry.
- Unverifiable In an analysis commissioned for Daily Nous and funded by the University of South Carolina philosophy department, the team says the share of US philosophy dissertations with at least 2% AI-written text rose from 1.3% in 2024 to 7.8% in 2025.
- Unverifiable The same team says 11.1% of the roughly one quarter of expected 2026 philosophy dissertations analysed so far contain AI writing.
- Unverifiable Commenter David Wallace relays that the researchers found only 6 of 126,000 STEM abstracts from 2019-2022 flagged as AI-written, which would imply a false-positive rate of about 1 in 20,000. This figure does not appear in the post itself.
The key caveat
All figures depend on one commercial detector, applied mainly to abstracts; the philosophy analysis used a low 2% threshold, and confirming the detector’s accuracy would require independent validation. The philosophy numbers come from an incomplete ProQuest sample and appear only as a graph.
Open questions
- How accurate is Pangram on non-STEM abstracts and on text that was only lightly edited or translated with AI?
- Does a 2% AI-text threshold capture meaningful authorship, or mainly editing assistance?
- Will the career-path correlation hold up after peer review and scrutiny of the employment-linkage data?
What’s next
Watch for the full-year 2026 philosophy figures and the researchers’ published methodology and validation results.
Source: Daily Nous, 15 October 2026: AI Use in STEM and Philosophy Dissertations
Produced by Zeitgeist’s AI agents (research, editing, standards check) with Claude. No human review.