Santiago Schnell has resigned as provost of Dartmouth College and is leaving the university, President Sian Leah Beilock announced Wednesday, after an independent faculty-led committee found he may have violated Dartmouth policies by using artificial intelligence in his published writing without disclosing it to the outlets involved. The New York Times and the Boston Globe first reported his departure this week. For more context on this story, see our ongoing latest AI developments
The resignation makes Schnell one of the most senior American academic administrators to step down over undisclosed AI use, and it follows months of reporting by student and independent outlets rather than a leak or a rumor. According to The Dartmouth, the college's student newspaper, whose investigation triggered the review, the finding rests on an eight-page committee report that examined his chat logs with Dartmouth's enterprise Claude account and tested his published prose with an AI-detection tool.
What the Committee Found
Beilock convened the committee last month to answer four questions, per The Dartmouth: whether Schnell met the AI disclosure requirements of the publications that ran his work, how he used AI in his academic and public writing, whether he was honest with those publications about it, and whether the conduct violated Dartmouth policies. The committee reviewed three published pieces, all submitted to outlets that either prohibited AI use or required its disclosure at the time, and limited its scope to writing produced since Schnell joined the college in July 2025.
Its conclusion was that Schnell may have violated Dartmouth policies by using AI in his public writing without proper disclosure. The committee also wrote that his cooperation with the investigation was "insufficient": although he turned over his Dartmouth-issued laptop, he did not provide logs of the personal LLM accounts he had used for academic and public writing, and he failed to make himself available at the requested times, according to the report as quoted by The Dartmouth.
An AI Detector Flagged the Writing
Central to the paper's reporting was Pangram, a commercial AI-detection tool. The Dartmouth's test of a September paper Schnell published in the Wiley journal ChemSystemsChem found it was "71% AI-written," while Semafor and The Dartmouth reported that an August 9 Washington Post column of his tested as "100% AI-written." The committee itself wrote that it did not use Pangram or other AI tools in its review, relying instead on Schnell's own chat history and publisher records.
The Washington Post Column
The committee found that Schnell's use of ChatGPT in the August 9 opinion column was a "clear violation" of the Washington Post's AI policies, The Dartmouth reported. The newspaper requires authors to confirm that submissions are "not created or manipulated with artificial intelligence or editing software," a spokesperson told Semafor for its August 26 investigation. Schnell acknowledged at the time that he had used the chatbot "to refine a few arguments, check for grammatical errors and submit it to the Washington Post."
Three Publications, Three Policy Breaches
The first paper the committee examined, "Reaction Acceleration: Reviving the second derivative in chemical kinetics," appeared in ChemSystemsChem in September. According to the report, Schnell used Dartmouth's enterprise Claude account to substantially edit the manuscript, troubleshoot code and generate figures, and asked Claude to check a version of a response he "asked ChatGPT to write." ChemSystemsChem has required authors to disclose AI use since at least 2023, the committee wrote, yet the published paper carries no such disclosure. When Schnell emailed the journal on October 6, he disclosed his ChatGPT use but not his Claude use, per The Dartmouth.
The third piece, on reproducible simulations of protein aggregation, was published September 19 in Springer's Bulletin of Mathematical Biology with two co-authors from the University of Notre Dame. Its policies at submission stated that any AI use "should be properly documented in the methods section of the manuscript," though the journal later relaxed them to allow undisclosed AI use for language and code refinement. The committee, which was chaired by former provost and computer science professor David Kotz and included Hany Farid, Odette Harris, Peter Orner and linguistics chair Lindsay Whaley, limited itself to cases where policies "could be readily ascertained."
What Happens Next
History professor Matthew Delmont will serve as interim provost through June 2028, effective immediately, Beilock announced. "I feel a sense of responsibility and accountability for the university, and I am proud to serve as interim provost," Delmont wrote in an email to The Dartmouth. Schnell declined to comment to the student paper.
Beyond one campus, the case is a test of whether disclosure policies at journals and newspapers have real consequences for powerful people. Publishers from Wiley to the Washington Post have spent two years writing AI rules; the Dartmouth review shows those rules being enforced against a member of a university's senior leadership, using his own chat logs as evidence. It also feeds a parallel debate about the reliability of AI detectors — the committee pointedly declined to rely on one — and suggests institutions will increasingly lean on account records and disclosure policies rather than detection scores alone.
---
Stay Ahead of AIGet the latest AI ethics news, analysis, and breakthroughs — all in one place.
Read more AI news →
