A university official writes a public op-ed warning about AI dishonesty in academia. Months later, an advisory committee finds that the same official failed to disclose significant AI use in that very op-ed, and in academic papers, despite publisher requirements to do so. The official resigns. This is not a hypothetical scenario designed to illustrate a point about hypocrisy. It is what happened at Dartmouth College, and it reveals something important about where higher education actually stands on AI disclosure, not where it says it stands.
The Gap Between What Was Said and What Was Done
Santiago Schnell, who served as Dartmouth’s provost, wrote a Washington Post op-ed in August 2026 titled “Universities are fighting AI cheating. But there’s a deeper problem.” The irony embedded in that title became apparent when an advisory committee concluded that Schnell had himself failed to disclose AI use in that piece, as well as in academic work, even where publishers explicitly required such disclosure.
Schnell’s described use of AI was not, by most measures, extreme. Logs from his Claude enterprise account, accessed through Dartmouth, show him using the tool to simulate peer review, asking the model to evaluate a paper as if it were a specialist referee. He also used it to revise arguments, check code, and edit figures. He told a journal editor he had used AI “as an editorial tool” to refine sentences and improve clarity. He told Semafor he had used chatbots to “refine a few arguments” and “check for grammatical errors.” His stated position was that he did not understand this kind of AI-assisted editing required disclosure at all.
The committee was not fully convinced. It noted that Schnell’s public statements may have understated the actual scope of his AI use, and that he “regularly went beyond his other public descriptions.” The investigation itself was complicated by what the committee called “insufficient cooperation”: Schnell did not provide logs from his personal LLM accounts, and he declined to meet with the committee.
One exchange from the Claude logs is particularly striking. Schnell uploaded a manuscript and asked the model: “Are there many signatures of AI writing in this article?” He later submitted a revised version that incorporated the kinds of changes Claude had flagged. Whether this reflects an attempt to conceal AI involvement or simply a desire to improve the text is something the committee could not determine with certainty. But the question itself is revealing.
A Thoughtful Framework, Inconsistently Applied
What makes this case genuinely complex is that Schnell was not simply dismissive of AI disclosure. In a September 2026 paper titled “Who Wrote This?”, he laid out a detailed and intellectually serious account of how he had used AI to produce that very paper, including conversations with GPT-5.6 Pro about the Turing imitation game, AI-generated early drafts, and Claude Fable 5.1 used for critical review before submission. He described checking cited literature himself, correcting errors, reorganizing arguments, and rewriting most sections. He was explicit: “I take full responsibility for the resulting manuscript.”
His theoretical position is coherent. He argued that the binary of “human-written” versus “AI-written” is too simple, that AI tools are closer to mathematical software than to co-authors, and that contribution and authorship are distinct roles. He resisted calling AI a co-author while insisting on full human accountability for the final text.
The problem is not the framework. The problem is that he applied it selectively. He disclosed extensively in one paper while failing to disclose at all in others, including the op-ed that was explicitly about AI integrity in academia.
What Disclosure Actually Means Now
Dartmouth’s president, Sian Leah Beilock, has been clear that she wants more AI at the university, not less. In announcing Schnell’s resignation, she wrote that leaders must “model the transparency and human judgment that we expect from our students and scholars.” The word she returned to was disclosure.
This is the real lesson, and it extends well beyond one university or one official. The question of what counts as AI use requiring disclosure is genuinely unsettled. Schnell’s own paper argues that the line between AI-assisted editing and AI-generated content is blurry and will only become blurrier. He is probably right. But that ambiguity does not dissolve the obligation to disclose: it makes disclosure more important, not less.
Survey data presented at Dartmouth’s own conference on AI adds another layer. Seventy percent of college students surveyed said that take-home work has become unfair because of AI, and 39 percent reported feeling pressured to use it because their peers do. Among Dartmouth students specifically, 92 percent expressed fear about AI’s effects. These are not students who are indifferent to the ethics of AI use. They are students watching their institutions navigate the same confusion they face, and looking to see whether the rules apply equally to everyone.
In Short
The Schnell case is not primarily a story about one person’s misconduct. It is a signal about the state of AI governance in institutions that are simultaneously promoting AI adoption and struggling to define what honest use looks like. Disclosure is the minimum standard everyone agrees on. The difficulty is that “disclosure” still means different things to different people, and the gap between stated policy and actual practice is wide enough for a provost to fall through.
Based on reporting from Ars Technica.
Read the original study.