Dartmouth provost resigns after review finds undisclosed AI use

Dartmouth President Sian Leah Beilock said Santiago Schnell resigned as provost and is leaving the university after an independent advisory committee concluded that he used AI to meaningfully alter published writing, failed to disclose that use and was not forthcoming about it. Beilock said in Dartmouth’s October 28 announcement that the decision to resign was Schnell’s.

Dartmouth named history professor and associate dean Matthew Delmont interim provost through June 2028. The personnel change followed a review Beilock commissioned on September 25 to assess Schnell’s AI use against the rules of publications that accepted his work.

The committee’s eight-page report identified at least two instances in which it said Schnell failed to meet publication requirements. For a sole-authored ChemSystemsChem paper, the committee said records from Dartmouth’s enterprise Claude account showed that Schnell used the system to substantially edit the manuscript, troubleshoot code and generate figures after peer reviewers requested major revisions. The report said the applicable policies required disclosure, but none was made when the paper was submitted or published.

For an opinion essay in a national newspaper, the committee said Schnell acknowledged using ChatGPT to refine arguments and check grammar. It concluded that the use clearly violated the newspaper’s policy barring submitted content created or manipulated with AI or editing software.

The committee said it did not use Pangram or any other AI tool to determine whether or how extensively Schnell used AI. Its evidence included Schnell’s communications and logs from Dartmouth’s enterprise Claude account.

The findings were limited in scope. Schnell had published at least 30 pieces since joining Dartmouth and listed seven more as forthcoming, but the committee conducted a targeted review under a compressed timetable. It said Schnell provided his Dartmouth-issued laptop but did not provide requested logs from personal large-language-model accounts, did not meet with the committee and did not propose another meeting time.

The committee examined a third paper involving AI-assisted review and editing of computer code but said the evidence did not clearly establish a publisher-policy violation. It also found no evidence that Schnell used AI to edit or manipulate that paper’s text.

On Dartmouth’s own rules, the committee said Schnell’s conduct could qualify as a violation of three institutional policies and concluded that his statements and disclosures fell demonstrably short of their letter and spirit. The committee stressed that it was not operating under those policies’ complaint procedures and did not recommend disciplinary action.