Dartmouth's provost used AI for his essay against AI cheating (and even his defense was AI)

By Boston Globe · September 25, 2026.

Santiago Schnell, provost (chief academic officer) of Dartmouth College for just over a year, published an op-ed in The Washington Post in August titled "Universities are fighting AI cheating. But there's a deeper problem." According to an analysis by the platform Semafor using the Pangram detector, published on August 26, the text was 100% written by AI. Weeks later, students at the university newspaper The Dartmouth extended the investigation to nine Schnell texts from 2026 and found a median of 96% AI-generated content, compared with 0% in his writings before 2022, according to Pangram.

The most awkward twist came when the provost tried to defend himself. The Dartmouth ran the statement Schnell sent by email to answer the accusations through the detector: it also came back 100% AI. That is, according to the student investigation, an academic leader accused of using AI to write about AI cheating used AI to draft his defense about his use of AI. Schnell maintains he merely used ChatGPT "to polish some arguments, fix the grammar and send it" after several rounds of his own editing, and notes that AI detectors "show bias against non-native English speakers" like himself, a Venezuelan.

The internal reaction has been harsh. Dartmouth's chapter of the American Association of University Professors (AAUP) is circulating a petition to remove Schnell for, according to his critics, violating the institution's Academic Honor Principle. "It's a textbook case of exactly what we tell our students not to do," said Christopher MacEvitt, professor and chair of the Religion department. Historian Bethany Moreton went further: "If we can't establish a humanistic experience in the classroom and intellectual guardrails that work at this moment, I don't know who we're asking to do that job for us." President Sian Leah Beilock announced on Friday a "rigorous and objective review" of Schnell's use of AI, after acknowledging that the rules on the matter are changing very fast.

The timing could not be more symbolic: Dartmouth is marking the 70th anniversary this year of the 1956 Dartmouth Summer Research Project, the gathering that gave the field of artificial intelligence its name, and it recently became the first Ivy League university to deploy AI "at institutional scale," together with Anthropic and Amazon Web Services. Beilock had published an article in The Atlantic just this month titled "Why I Want More AI at Dartmouth." The school's own policy expressly bars students from using generative AI tools in coursework without explicit authorization, which deepens the perception of a double standard between what is demanded of students and what is tolerated at the top.

Beyond the specific scandal, the case touches a nerve almost no university has resolved: what counts as authorship when a text can be polished, rewritten or even almost entirely generated with AI without the result ceasing to "sound" human. Dorothy Leidner, a professor of AI ethics at the University of Virginia, directly disputes Schnell's argument about detector bias: she notes that AI produces grammatically flawless English, so one would expect the opposite effect in texts by non-native speakers. For Leidner, the underlying question is not technical but philosophical: "writing is only human when it faithfully represents a person's thinking," and shaping ideas into words is itself part of authorship.

Our reading is that this episode is less a campus anecdote than a preview of a problem every text-based institution —academic, editorial, corporate— will have to solve in the coming years: the gap between the technical ability to generate prose indistinguishable from human writing and the near-total absence of clear disclosure rules applied equally at every rung of the hierarchy. Detectors like Pangram, with their claimed near-zero false-positive rate, have become de facto arbiters of a debate no university has settled in writing, and that creates shaky ground where the penalty depends more on who points the finger at you than on a previously agreed rule. In the short term, the cost is trust: institutions that demand academic honesty from their students while their top leaders use the same tools without disclosing it feed exactly the cynicism that is part of the price of this transition.

In the medium term, however, it is foreseeable that cases like this —embarrassing but clarifying— will accelerate what should have existed from the start: explicit, symmetrical disclosure policies, applying equally to a first-year student and to the provost. That is, in fact, the usual pattern of disruptive technologies: first the gray area and the scandal, then the rule. And if AI's long term involves freeing people from the mechanical part of intellectual work —including much of the routine writing— so they can devote their judgment to what really matters, the question Dartmouth leaves behind is not whether AI should be used to write, but who has the moral authority to set that rule when they fail to follow it even while defending themselves against having broken it.