Santiago Schnell Resigns as Dartmouth Provost Over AI Use
Santiago Schnell resigned as Dartmouth provost after a faculty-led investigation found he may have violated university policies by using AI in some academic and popular writing without proper disclosure. Dartmouth President Sian Leah Beilock said Schnell is leaving the university; an interim provost will serve through 2028. Catholic scholars expressed surprise at the resignation, with some saying the case highlights uncertainty over AI use in higher education. The case has renewed debate about rapidly expanding AI use in professional writing and the standards for disclosure.
5 hours ago
Santiago Schnell resigned as Dartmouth’s provost after a faculty-led investigation found that he had used AI in some academic and public writing without required disclosure. The case has prompted debate among Catholic scholars about accountability and unclear standards for AI use in higher education.
Dartmouth President Sian Leah Beilock announced that Schnell was resigning and leaving the university. She said the college would have an interim provost through 2028.
An independent, faculty-led committee concluded that Schnell had used AI to “meaningfully alter” writing in ways that breached at least one publication’s AI-use or disclosure requirements. The committee also found that he failed to disclose AI use in violation of a national newspaper’s policy, was not forthcoming about that use, and fell short of standards in academic codes of conduct.
Scrutiny intensified after The Dartmouth reported in September that an AI-detection tool had flagged several of Schnell’s writings dating to 2022, including academic papers and opinion pieces, as overwhelmingly AI-written. The report followed a Semafor investigation that had flagged a Washington Post opinion article by Schnell as “100%” AI-generated.
Schnell said he used AI for language refinement, copyediting, clarity, organization, and efficiency, while retaining responsibility for the final text. He said he developed the ideas, conducted the research, evaluated evidence, prepared drafts, and determined what appeared in the final version; he also questioned the reliability of AI detectors and said they can be biased against nonnative English speakers.
Some Catholic scholars and higher-education reform advocates expressed surprise at the resignation. Jennifer Frey, a University of Virginia philosopher who opposes AI-written work, said the decision gave her “whiplash,” pointing to Dartmouth’s encouragement of AI use and the lack of settled norms in higher education.
Classical Learning Test founder Jeremy Tate called Schnell’s departure a major loss and questioned whether he had received extra scrutiny, while acknowledging that Schnell may have made a mistake. Frey said Dartmouth could have used the case to prompt a transparent discussion about acceptable AI use before imposing discipline.
Schnell, a mathematical biologist and former University of Notre Dame science dean, had been appointed Dartmouth’s chief academic officer for the 2025–26 academic year. He was also known among Catholic higher-education advocates for promoting a classical vision of university education and urging U.S. bishops to press Catholic universities to uphold their religious mission.
Human authorship entails moral responsibility for AI-assisted scholarship
Human authorship entails moral responsibility for AI-assisted scholarship because AI is not a moral agent: it can generate or organize material, but it cannot answer morally for the claims it helps produce. The human scholar remains responsible for how AI is used, what is accepted or rejected, and what is ultimately presented to readers. That responsibility is shared with developers and institutions in relevant ways, but it cannot be transferred to the tool.
The Church’s 2025 note Antiqua et Nova distinguishes a machine’s operation from a person’s moral agency. A system may select among outputs by rules or statistical inference, but a human being can deliberate about truth and good, act according to conscience, and accept responsibility for a decision. Applied to scholarship, this means that generating text is not the same as exercising scholarly judgment. A system can propose a quotation, explanation, or argument; the scholar must decide whether it is accurate, relevant, fairly represented, and appropriate to publish.
The note states the principle directly:
“ultimate responsibility for decisions made using AI rests with the human decision-makers”
In scholarship, the “decision” is not only the final decision to submit or publish. It includes choosing the research question, deciding what tasks to delegate, judging the outputs, and incorporating them into an argument. A scholar who relies on AI without checking its claims cannot make the system responsible for the resulting errors. The scholar chose to use and rely on it, and therefore remains accountable for that choice.
Responsibility is not limited to whether a paper contains a factual mistake. Antiqua et Nova stresses that AI should be directed toward human dignity and the good, and that ethical judgment must consider how a technology is used and toward what ends. It also describes responsibility as care for others, not merely accounting for results.
For scholarship, that broader responsibility includes the effects of one’s work on readers, research participants, colleagues, and the academic community. A scholar should ask not only, “Did the tool produce a correct sentence?” but also: “Does using it in this way serve honest inquiry? Am I presenting its contribution in a way that could mislead readers about how the work was done? Have I protected information entrusted to me?” The supplied Church texts establish the broader ethical principles, though they do not prescribe specific academic rules for disclosure, citation, or data handling. Those details may depend on the relevant journal, university, discipline, and applicable law.
Calling an AI system an “author” can obscure a crucial distinction. In the Church’s account, moral responsibility belongs to persons, not artificial systems. Thus, even if a tool makes a substantial contribution to wording or analysis, it cannot bear the moral duties associated with authorship: standing behind the work, answering for its claims, or correcting the record.
That does not settle every institutional question about authorship credit or intellectual property. The supplied sources do not define academic authorship or establish a universal policy for naming AI tools in a paper. But they do support a clear moral conclusion: whatever formal credit a system receives under an institution’s rules, responsibility for using its output and presenting the resulting work remains with human persons.
Nor does this mean that the researcher is the only party with responsibility. The Church’s note says responsibility begins with those who develop, produce, manage, and oversee AI, and is also shared by those who use it. Developers and institutions may bear responsibility for system design, safeguards, or oversight. But shared responsibility is not an excuse for the author to disclaim responsibility for the scholarly work they choose to submit.
A responsible scholar should treat AI output as material requiring judgment, not as an authority whose fluency guarantees truth. The Church warns that complex AI processes can be difficult to understand and that opacity makes accountability harder; it therefore emphasizes responsibility at each stage of AI-assisted decision-making. In practice, this supports habits such as checking claims against reliable sources, confirming quotations and references, reviewing generated analysis for distortion, and retaining enough understanding of the research process to explain and defend the published conclusions.
The scholar should also avoid becoming so dependent on the tool that personal judgment is displaced. Antiqua et Nova cautions users against excessive reliance on AI for decision-making. In scholarship, that caution is especially relevant when a tool is used to summarize literature, interpret evidence, or draft arguments: the scholar must still understand the sources and reasoning well enough to take responsibility for them.
The thesis, then, is sound if understood carefully: human authorship entails accountability for the human choices that shape AI-assisted work. It does not imply that every error automatically proves the same degree of personal fault, nor that scholars alone are responsible for every consequence of a complex system. It does mean that an author cannot use AI as a moral shield. The Church’s governing principle is that technological assistance may be delegated, but human moral responsibility cannot.