A Provost Resigned Over AI Disclosure. The Committee Never Ran a Detector.

The committee reviewing Dartmouth’s provost declined to run an AI detector on his writing. A month earlier, that same college cleared that detector for faculty to use on student work.

President Sian Leah Beilock announced on October 28 that Provost Santiago Schnell had resigned, one day after the date on the committee’s report. Most of the coverage stopped at the irony, since he had written a Washington Post column about AI cheating. The more useful part is the report itself, all eight pages. It shows what a faculty committee treated as evidence of AI use when the person under review was the college’s chief academic officer. You can copy that list this week, for your own AI disclosure first and for your next student case second.

What the Dartmouth report found on AI disclosure

The committee had five members, four Dartmouth professors and a trustee, chaired by computer scientist and former provost David Kotz. Beilock asked for a review “measured against the criteria of the publications to which he submitted his work.” The committee found that he “used AI to meaningfully alter his writing in violation of at least one publication’s AI use and/or disclosure requirement, failed to disclose his use of AI in clear violation of a national newspaper’s stated policy, was not forthcoming about his AI use,” and fell short of three Dartmouth conduct policies.

Three limits belong next to that sentence. The report says the committee was not “recommending any related disciplinary action.” Beilock wrote that the resignation “was his decision.” And on a third paper the committee said it “could not conclude there was a violation,” because the evidence showed AI reviewing code while the publisher’s policy at that stage did not clearly require disclosure. That restraint is part of the lesson.

What the committee read instead of a score

The report’s first paragraph settles the detector question: “The Committee did not use Pangram or other AI tools when determining whether and to what extent Provost Schnell used AI.” Kotz told The Dartmouth the committee felt detector results did not constitute “evidence” of AI use. “We did not feel it was necessary or appropriate to use a technical tool to determine whether he used AI,” he said.

Now hold that next to September 29. That day the interim dean of the faculty of Arts and Sciences sent undergraduate faculty guidelines permitting them to submit student writing to Pangram, the one detector authorized, “because it has shown the most reliability of available detectors.” Those guidelines are more careful than the headline, as we covered in October. A score alone cannot carry a finding. Still, a student at that college could fairly ask why a tool is allowed near her paper when the provost’s reviewers would not use one on his.

So what did the committee read?

• Logs of his use of the college’s enterprise AI account, Claude, prompts included.
• His own email, to the president, to co-authors and to a journal.
• What he told Semafor.
• Publisher policies as they stood at submission and at publication, including archived copies captured on the web.
• The published work itself, and whether a disclosure appeared in it.

Every item on that list is a record made at the time. None of it is a guess about prose.

The gap between the description and the log

Read closely, and the findings keep returning to one thing: the distance between what someone said he did and what the record shows.

On a chemistry paper, Schnell told the president he had relied on older author instructions. The committee found the publisher “has been requiring the disclosure of the use of AI since at least 2023.” The logs include prompts such as “Please make the corrections to the files, and the repo. If you have to generate new figures, please provide them to me.” When he later wrote to the journal, his note named ChatGPT and did not name Claude.

On a forthcoming article, the logs show him asking, “Are there many signatures of AI writing in this article?” The report says he then submitted a revised manuscript that “incorporated the kinds of changes Claude suggested.” When an editor asked, he described AI as an editorial tool to “help refine sentences, improve clarity, and the expression of arguments.” The committee wrote that this statement “understates the nature of the Provost’s actual AI use.”

Here is the part that should bother anyone who publishes. The rule over your work may be broader than you assume. The Washington Post policy quoted in the report had writers agree their content “was not created or manipulated with artificial intelligence or editing software.” Springer Nature now sorts AI use by risk, and its lowest tier covers grammar, language editing and translation. The declaration for that tier still reads, “Clearly describe use and confirm author accountability.” The same page says, “Declaration does not negatively influence editorial decisions.”

If you have not read those sentences for the venues you publish in, that is the fix, and it costs about twenty minutes.

The case against all of this

Plenty of smart people think this ended badly, and you should hear them. Schnell’s attorney told The Washington Post, as The Dartmouth reported, that he did not have a “meaningful” chance to respond. Kotz himself said “the most important thing” the committee missed was that they “were never able to talk to the provost himself,” and the report says he did not make himself available or hand over logs from personal AI accounts. Reed Albergotti, the Semafor reporter whose analysis started this, wrote that the resignation “wasn’t the right call” and told The Dartmouth it was an “unfair singling out of one person.” Professor Eugene Demidenko called the outcome “too harsh,” because disclosure “was never explained by our administration.”

Others land elsewhere. Professor Molly Geidel said, “we certainly shouldn’t tell students that they have to do that work themselves when we are not doing it ourselves.” Student senator Jason Zhu said the open question is “how we address the discrepancy between students and faculty.” Kotz is due to speak to the general faculty on November 9, so more may come. This post cannot tell you whether the resignation was right. The disclosure problem is yours either way.

Run the same test on your next student case

When a flagged paper lands on your desk, ask what you would have without the score. A dated outline. A draft history. Notes from a conversation. Those are records made at the time, the same kind the committee went looking for.

Then do the thing Kotz named as the biggest miss, and talk to the person. Dartmouth’s faculty guidelines already require communicating with a student before a detector result affects a grade, and they promise a “constellation” of “information beyond the result.” We saw the same pattern in September, when mathematicians argued over a proof with receipts, not detectors.

Build your AI disclosure receipt this week

Pick one piece of writing that is still in progress, then do six things.

1. Name the piece and the venue. A manuscript, a column, a grant section. Something with your name on it and someone’s rules over it.
2. Copy the venue’s AI policy word for word, with today’s date. If it changed since you started, copy both versions. The instructions you remember from your last submission do not count.
3. List what AI touched. Text, code, figures, translation, references, critique. Name the tool and the account, personal accounts included.
4. Save the record. Keep your drafts, and keep the prompts that changed an argument.
5. Write the disclosure sentence in the venue’s own terms. Plain verbs. Say what the tool did.
6. Read it with your log open. If a stranger holding that log would say your sentence understates, rewrite it. Then show your co-authors. One paper in this report had a graduate student’s name on it, and that student carries your choices too.

What this will not do

Straight talk. No study tests a receipt like this, and it is a judgment call, not legal advice. It will not rescue a use your venue bars outright, as the Post’s language appears to. When a policy is unclear, ask the editor in writing and keep the answer. It will not settle whether AI belongs in scholarly writing at all. Reasonable colleagues disagree, and your discipline may draw that line harder than your publisher does.

Start with one piece of writing

The free session walks you through all six moves, and you leave with a finished receipt.

Then the next thought arrives on its own. That is one piece of writing. You also have syllabi, rubrics and announcements, and each syllabus carries its own aging AI language. The Course Refresh Assistant takes an old syllabus and gives you a live Course Record in minutes: your course, your outcomes, your policies and your voice, in the one place every tool in the toolkit reads from. It will not keep your receipts for you. Full access for 14 days, cancel in one click. Write this first receipt by hand anyway, because the sentence has to be true in your own words.

If you would rather compare notes with other instructors, the Harvest Kernel Learning Community is free, and today’s question there is simple: when did you last read the AI policy of the journal you publish in most?

A detector guesses at your prose. A record shows your choices. Keep the record before anyone asks for it.

Sources

• Advisory Committee, memorandum to President Sian Leah Beilock, “Review of Concerns Regarding AI Use by Provost Santiago Schnell,” October 27, 2026. president.dartmouth.edu
• Sian Leah Beilock, “An Update on the Provost Review,” Office of the President, Dartmouth, October 28, 2026. president.dartmouth.edu
• Sian Leah Beilock, “A Review and a Community Conversation on AI,” Office of the President, Dartmouth, September 25, 2026. president.dartmouth.edu
• Alex Klee, Brennan Kelly, Jared Gonzalez Arce and Veda Dua, “Schnell’s resignation prompts conversation on appropriate consequences for AI use,” The Dartmouth, November 6, 2026. thedartmouth.com
• Jackson Hyde, “Dean of faculty authorizes professors to use AI detector Pangram to review student work,” The Dartmouth, October 1, 2026. thedartmouth.com
• Reed Albergotti, “My reporting prompted Dartmouth provost to resign. It wasn’t the right call,” Semafor, October 30, 2026. semafor.com
• Heather Hollingsworth, “Dartmouth provost resigns following accusations he used artificial intelligence for writing,” Associated Press via ABC News, October 29, 2026. abcnews.com
• Springer Nature, “AI use in manuscript preparation,” read November 6, 2026. springernature.com