Dartmouth opens review of provost’s AI-assisted writing

Dartmouth College has begun an independent review of Provost Santiago Schnell’s use of artificial intelligence after students and faculty called for him to be fired or resign over AI-assisted professional writing. The review will assess his use of the technology against the standards of publications that received his work. President Sian Leah Beilock said the inquiry was warranted because Schnell, as the college’s chief academic officer, represents Dartmouth and senior leaders are expected to meet the highest standards. The advisory committee will include faculty and alumni, and Dartmouth also plans campus conversations on AI, authorship and transparency. The controversy grew from a Semafor investigation published in August that used the Pangram detector to assess opinion pieces. It identified Schnell’s Aug. 9 Washington Post column as 100% AI-written. Pangram labels material as AI-generated when it finds that more than 80% of the text was produced by AI. Schnell said he wrote the essay from his own thinking, then used ChatGPT to refine arguments, check grammar and submit it to The Washington Post. The newspaper’s guest-writer policy requires contributors to confirm that submissions were not created or manipulated with artificial intelligence or editing software. The Dartmouth subsequently tested Schnell’s published work and found a median Pangram score of 96% AI-written across nine academic articles and opinion pieces published in 2026. By contrast, works published before ChatGPT’s public release were classified as 100% human-written. The newspaper also found that a Sept. 18 email from Schnell to The Dartmouth was classified by Pangram as 100% AI-written. Schnell said he develops the ideas and arguments, conducts research, evaluates sources and evidence, determines the structure, writes drafts and makes the final decisions. He described AI as a tool for language refinement, copyediting, clarity, organization and efficiency, and said he reviews the final text and takes responsibility for it. He declined to provide AI logs to The Dartmouth, arguing that a detector’s classification is not a determination of authorship and that the relevant issue is the role a tool played in producing a work. The published detector results varied across individual articles. Pangram rated his article on AI-text detection, “Who Wrote This? Turing, Total Variation and the Mathematics of AI-Text Detection,” at 99% AI-written, and “From Signals to Measurands” at 92%. Schnell acknowledged using ChatGPT for language refinement and Python-code review in the latter paper. The results do not independently establish how the writing was produced. A 2023 Stanford study found that common detectors consistently misclassified writing by nonnative English speakers as AI-generated, while an August preprint from the University of Notre Dame found that detectors including Pangram 3.2 could not distinguish AI editing from a complete large-language-model draft. Pangram has reported a 0.01% false-positive rate. Dartmouth’s Academic Honor Policy requires students to acknowledge generative-AI use in coursework, while course-specific rules are set by instructors. The college’s own publishing guidance says generative AI can accelerate editing but also raises questions of authorship and research integrity. Eli Moyse, writing in The Dartmouth, argued that Schnell should resign because the alleged extent of his AI use conflicted with Dartmouth’s academic standards. He said that presenting work produced by AI as one’s own could constitute plagiarism. The dispute comes as Dartmouth has encouraged the adoption of generative AI. Its 2025 AI report recommended AI literacy across all degree programs beginning in the 2026 academic year and urged a corporate partnership with a major foundation-model provider. Beilock said the episode showed the need for rules suited to different roles, describing leadership on a rapidly changing technology as inherently likely to involve missteps. The issue also reflects broader use of AI by university leaders. Inside Higher Ed’s 2026 survey found that 65% of chief academic officers used AI to draft communications to faculty, staff or students, and 71% used it for work at least weekly. Dartmouth’s review and planned community discussions will focus on how AI use should be disclosed and judged in academic leadership and publishing.