Dartmouth College's provost is facing calls for his removal after an AI-detection tool flagged an opinion essay he wrote about fighting AI cheating as itself being entirely written by artificial intelligence. Santiago Schnell, who serves as the Ivy League school's top academic officer and is a professor of mathematics and adjunct professor of biochemistry and cell biology and biomedical data science, published a Washington Post guest column on Aug. 9 titled “Universities are fighting AI cheating. But there's a deeper problem.” Now faculty and students say the piece — and several others bearing his name — may not have been his own work at all.
The detection company Pangram identified Schnell's Washington Post piece as 100 percent AI-written, according to Semafor's analysis, as reported by aicommission.org. That same Semafor review found that 10 of 310 guest submissions to The Wall Street Journal, The Washington Post, and The New York Times over the prior month were labeled at least 80 percent AI-generated, with another 40 articles flagged as partially AI-generated. Schnell's essay stood out as one of the clearest cases in the batch.
Student newspaper The Dartmouth went further, publishing an exposé examining several 2026 opinion pieces and academic articles bearing Schnell's byline. Using Pangram 4.0, the paper reported that nine Schnell works published in 2026 carried a median AI-written percentage of 96 percent — while Pangram 4.0 labeled all of his written works before 2022 as 100 percent human-written, according to The Boston Globe.
Schnell Defends His Process, Apologizes for Not Disclosing It
Schnell has pushed back on the idea that the machine wrote his work for him. He said he wrote the Washington Post essay based on his own thinking, then used ChatGPT to refine a few arguments, check for grammatical errors, and submit it — a process detailed in the same Semafor-sourced account carried by aicommission.org. He has separately maintained that AI tools are assistive tools that do not replace his research, judgment, or authorship, per the Globe's reporting.
In a Sept. 18 emailed statement to The Dartmouth explaining his process, Schnell said he uses AI to refine writing and improve efficiency. But The Dartmouth reported that statement itself registered as 100 percent AI-written when it was run through the detection software — a detail that has only deepened skepticism on campus. Schnell has since said he should explicitly disclose AI use in public writing going forward, and said he was sorry that doing so was not already his standard practice.
Schnell has also argued that AI detectors can be biased against nonnative English speakers, including himself, according to the Semafor-derived reporting. That argument echoes a real fault line in the research: a July 2023 Stanford study found that seven tested AI detectors flagged more than 60 percent of human ESL writing samples as AI-generated, though Pangram was not among the tools evaluated in that study, according to Pangram. An August 2024 study by ETS, the testing organization, analyzed roughly 2,000 writing samples from nonnative English speakers and found no bias in its own detectors, though it did not evaluate commercial tools like Pangram.
Pangram Says Its Own Testing Shows Low Bias
Pangram itself reports a different picture for its current system. The company says it recorded an overall false-positive rate of just 0.078 percent on four public ESL datasets, compared with its stated general false-positive rate of 0.01 percent, and claims its tool is two orders of magnitude more accurate than TurnItIn on ESL text. On the widely cited Liang TOEFL dataset, Pangram says it produced no false positives, while GPTZero self-reported a 1.1 percent false-positive rate on that same study.
Separately, Pangram's underlying technology — described in an academic paper as a transformer-based neural network trained to distinguish text written by large language models from text written by humans — reports error rates more than 38 times lower than zero-shot methods like DetectGPT and other leading commercial detectors, based on a benchmark of 1,976 documents spanning 10 text domains and outputs from eight large language models, according to the paper posted on arxiv.org. That same paper notes that commercial detectors such as TurnItIn have shown high false-positive rates in academic-plagiarism settings, and that zero-shot detection methods generally have been shown to be biased against nonnative English writing.
Faculty Call Schnell's Conduct a Violation of the Honor Code
On campus, the reaction has moved from scrutiny to demands for accountability. The Dartmouth chapter of the American Association of University Professors has circulated a petition calling for Schnell's removal, and Dartmouth faculty have accused him of violating the college's academic honor principle, according to the Globe's reporting. Some faculty have called outright for his firing.
Dartmouth community members, per the Globe, see hypocrisy in leaders' ethical AI statements and practices.
Dartmouth President Sian Leah Beilock has announced a rigorous, objective review of how Schnell has used AI, and said the college needs role-appropriate rules and policies for the technology. Beilock, who has separately authored an Atlantic piece titled “Why I Want More AI at Dartmouth,” has framed the episode as evidence that Dartmouth needs clearer guardrails even as it embraces the technology broadly.
A Provost's AI Use Collides With Dartmouth's AI Legacy
The controversy lands at an awkward moment for the college, which is marking its 70th anniversary.
The episode has prompted broader questions about AI ethics and accountability among academic leaders, according to the Globe.
For now, Schnell remains in his post as the review Beilock ordered gets underway, with no timeline yet made public for when it might conclude.
