What happened at Dartmouth on October 28?
Dartmouth College provost Santiago Schnell resigned on October 28, 2026, and is leaving the university. President Sian Leah Beilock announced the decision in a message to the campus community. She said the resignation was his own decision.
The announcement followed an eight-page report from an independent faculty committee. The committee found that Schnell "used AI to meaningfully alter his writing" in breach of at least one publisher's rules. It also found that he "was not forthcoming about his AI use."
Beilock wrote that she expects senior leaders to meet the highest standards. "Out of respect for his colleagues and the institution, Santiago has resigned as provost and is leaving the university," she wrote.
Matthew Delmont, a history professor and associate dean of faculty, became interim provost immediately. He will serve through June 2028. Delmont has written six books, including Half American, which won the Anisfield-Wolf Book Award.
The resignation came less than two weeks before faculty leaders planned a no-confidence vote, Inside Higher Ed reported. More than 220 faculty, staff, students and alumni had signed a petition organized by Dartmouth's AAUP chapter seeking his removal, according to The Dartmouth.
The dispute began with an opinion piece. On August 9, 2026, The Washington Post published Schnell's guest column arguing that universities were not doing enough about AI cheating in student work.
Within three weeks, that column was at the centre of a national argument about AI on the opinion pages, and about who enforces the rules.
What did the review committee actually find?
Beilock convened the committee on September 25, 2026. David Kotz, a computer science professor and former Dartmouth provost, chaired it. The other members were Hany Farid, Odette Harris, Peter Orner and Lindsay Whaley. The committee delivered its report on October 27.
The review was broad in scope but partial in practice. Schnell had published at least 30 writings since becoming provost in July 2025, with seven more listed as forthcoming. The committee examined three published pieces and one forthcoming article.
The charge had four parts. The committee was asked whether he met publishers' AI rules, how he used AI, whether he was honest, and whether his conduct violated Dartmouth policy. It answered each question, while noting that a full review of his work was impossible in the time allowed.
The first case involved a paper in ChemSystemsChem, published September 16, 2026. Records from Dartmouth's enterprise Claude account showed Schnell using the tool to edit the manuscript after peer review. He also used it to troubleshoot code and generate figures. One prompt asked Claude whether it found "any scientific issue with the paper".
The journal required authors to disclose that kind of AI use. Schnell provided no disclosure at submission or publication. He told Beilock he had relied on pre-2025 author instructions. The committee found the publisher had required disclosure since at least 2023. On October 6 he emailed the journal about ChatGPT use, but did not mention Claude.
The second case was his Washington Post guest column. A newspaper spokesperson told Semafor that writers must confirm submissions were "not created or manipulated with artificial intelligence or editing software." Schnell told Semafor he used ChatGPT "to refine a few arguments". The committee called this a clear violation of the stated policy.
The third paper, in the Bulletin of Mathematical Biology, produced no finding of violation. The journal changed its AI policy during revision, and the committee found no evidence that Schnell used AI on the paper's text. He had told co-authors that he ran their code through Claude and ChatGPT.
A forthcoming article in the Journal of Catholic Higher Education raised separate honesty questions. Schnell uploaded the manuscript to Claude and asked, "Are there many signatures of AI writing in this article?" He later submitted a revision that incorporated suggested changes, the report says. He told the editor he had used AI as an "editorial tool". The committee said that description understated his use.
The committee also faulted his cooperation. Schnell supplied his Dartmouth laptop but withheld logs from personal AI accounts. He did not make himself available to discuss his AI use at the requested time, the report says. The committee warned that its findings might have changed with more information.
Its final judgment was careful. The committee said his conduct "could qualify" as a violation of Dartmouth policies, and it did not recommend disciplinary action. It described the limits of its own evidence, including the missing records.
The report described failures in "at least two instances". It also found that his public statements about his AI use were incomplete. That distinction matters: the case is about disclosure and candor, not only about using ChatGPT.
Did Santiago Schnell admit to using AI?
Schnell acknowledged using AI tools for editing and language work from the start of the dispute. He denied that AI replaced his research, judgment or authorship.
In a September 18 email to The Dartmouth, he said he used AI for "language refinement, copyediting, improving clarity and organization". He added: "I review the final text and take responsibility for its content."
A week later he published a letter to the editor in the student newspaper. "In retrospect, I now realize I should always explicitly disclose use of AI in my public writing," he wrote. He apologized and said that would be his standard practice going forward.
His explanations shifted as the evidence grew. On September 10 he called the "100% AI-written" label attached to his column inaccurate. He said he wrote the essay himself and used ChatGPT to check arguments and grammar. He later told Beilock he had relied on outdated journal instructions. The committee found no support for that account.
The column's subject sharpened the reaction. Schnell wrote that a degree "should distinguish what students can do independently from what they can accomplish with AI". The committee later found that he had not disclosed the help he received on that same essay.
The report does not accuse him of fabricating research. Its objection was narrower. It said his disclosures, and his public descriptions of his process, fell short of what publishers and Dartmouth required.
He declined to comment to The Dartmouth when his resignation was reported. No public rebuttal from him appeared in the coverage reviewed for this article.
How did an AI detector start the case, and can detectors be trusted?
The story began with a detector, not a whistleblower. On August 26, 2026, Semafor published an analysis of guest opinion submissions at three national newspapers. It used Pangram, a commercial AI-text detector.
Semafor examined 310 guest submissions over a month at the Wall Street Journal, the Washington Post and the New York Times. It labelled 10 as at least 80% AI-generated and another 40 as partly AI-generated. Schnell's column scored 100%.
That count covered a single month, so it was a snapshot rather than an audit of the whole industry. It was enough to prompt questions about how closely newspapers check guest work.
The analysis followed a separate controversy over an op-ed by investor Stanley Druckenmiller. He told NOTUS, as quoted by Semafor: "I write everything using AI now for the same reason I use a calculator when I do math problems." The Journal's opinion editor, Paul Gigot, defended publishing it.
The Dartmouth then ran its own Pangram tests on Schnell's output. Nine pieces published in 2026 had a median score of 96% "AI-written". His work from before 2022 scored as fully human. Even his emailed statement to the paper was flagged as 100% AI-written.
Detectors remain contested. Pangram says its false-positive rate is 0.01%, and a University of Chicago audit in October 2025 found a near-zero error rate, according to The Dartmouth. Critics say such tools can misclassify polished or formulaic prose. A score describes patterns, not authorship.
The committee avoided that dispute. It did not use Pangram or any AI detector, the report states. Instead it compared specific submissions with publisher rules and reviewed Schnell's own chat logs. The approach ties findings to disclosure obligations rather than to a statistical score.
The limit is that the record was incomplete. Without logs from his personal accounts, the committee could not measure the full extent of his use. Documents, not detector scores, decided the case.
What do Dartmouth's own rules say about AI?
Dartmouth requires students to disclose AI use rather than banning the tools. The Academic Honor Principle is the rule that applies in coursework. The Office of the Provost is listed as the office of primary responsibility for that principle.
The committee applied three policies. They were the Code of Ethical Conduct, the Academic Honor Principle and a Geisel School of Medicine professionalism policy. The Geisel policy explicitly covers op-eds as well as scholarship.
Beilock has argued that the institution needs clearer rules. "We need role-appropriate rules and policies for this technology," she wrote in September. She also said that leading on a new technology is messy and will involve missteps.
Campus debate has been active. Nearly 400 people filled the Grand Ballroom at the Hanover Inn this month to discuss AI in higher education. Dartmouth also hosts a conference this week called AI at 70: What Must Remain Human? It marks 70 years since the 1956 summer workshop credited with launching the field.
Why does one provost's writing matter beyond Dartmouth?
The case lands on a campus that calls itself the birthplace of artificial intelligence. Dartmouth is marking that history this week, even as its chief academic officer leaves over how he used the technology in his own work.
The scale is unusual. A provost published more than 30 writings in about 15 months, a volume that makes individual disclosure hard to track. The committee reviewed a fraction of them and found violations in two. It could not establish the full picture.
Publishers are affected too. Their rules differ, and enforcement is uneven. The Washington Post's policy was clear and appeared in a competitor's story before it produced any visible consequence. Some require disclosure in a methods section; others have no specific AI policy at all.
For faculty, the precedent cuts both ways. A chief academic officer was held to the same integrity standard as students. That answers a complaint repeated on campus, where professors questioned whether rules applied equally to administrators.
Individual professors had argued for that consistency for weeks. Biological sciences chair Magdalena Bezanilla told The Dartmouth that any use of generative AI should be disclosed along with its extent. Economics professor Paul Novosad said there was "certainly not a shared understanding" of responsible use among faculty and administrators.
For employees, the review of enterprise Claude logs raises a privacy question. Dartmouth examined records from a work account during an investigation. Employees who use employer-provided AI tools may find those conversations can be reviewed later.
For other universities, the case offers a template. The committee did not rely on detection scores or public outrage. It compared specific submissions with the rules in force at the time. It documented gaps and left discipline to the president.
The limits matter too. The findings rest on partial records and a short deadline. The committee was asked for answers by October 21. He did not provide logs from personal accounts when asked, and the committee said so plainly.
Students and faculty pushed the issue before the report existed. The AAUP petition and the planned no-confidence vote showed that governance pressure, as much as any detection score, drove the decision. That pressure will shape how other campuses handle similar cases.
The practical lesson for institutions is to keep records. Chat logs and draft histories gave this committee more purchase than a percentage score. Universities that want to enforce disclosure rules may need policies on retaining and auditing AI conversations, with clear notice to staff.
What happens next at Dartmouth?
Delmont takes over the provost's office at a busy moment. The interim appointment runs through June 2028, which gives Dartmouth time before a permanent hire. Neither Beilock's message nor the report described a search process.
The faculty vote that was meant to test Schnell's support is now moot. The debate about AI rules is not. Faculty, students and trustees are still working out what disclosure should look like in classrooms, journals and opinion pages.
Wiley, the Bulletin of Mathematical Biology and the Journal of Catholic Higher Education did not immediately respond to questions from The Dartmouth.
Three things are worth watching. Whether other universities adopt explicit AI-disclosure rules for faculty. Whether journals start checking disclosures more aggressively. Whether Dartmouth's conference produces concrete policy rather than another conversation.
The provost's job is now held by a historian of African American life and the civil rights movement. He inherits a faculty debate that the report resolved for one person, not for the institution.