Dartmouth’s provost warned us about AI cheating—but his own AI use cost him his job

Regardless of the scandal, Dartmouth President Sian Leah Beilock desires more AI. (She composed a piece for The Atlantic in September called “Why I Want More AI at Dartmouth.”) What’s important, however, is that the school’s leaders need to “hold themselves to the greatest requirements,” she stated. This appears to indicate one main point: disclosure.

In revealing Schnell’s resignationBeilock composed that “each people brings a various viewpoint on how and when to integrate AI into our work and our lives. Keeping Dartmouth at the leading edge of AI undoubtedly indicates leading through an environment that is fluid and fast-changing. That is why, as leaders, it is so vital that we design the openness and human judgment that we anticipate from our trainees and scholars.”

Who composed this?

What does “disclosure” imply? In a September 2026 paper on AI detection tools, called “Who Wrote This?”, Schnell himself goes to terrific lengths to argue that “human-written” or “AI-written” is far too easy a binary. In his view, whatever will end up being a mix of human judgment and AI work, and there’s absolutely nothing incorrect with that. The secret for him is that the human author takes complete duty for the text, nevertheless it was made up and modified.

In the paper, Schnell sets out his approach for producing it. It’s worth pricing estimate at some length:

In September 2026, I asked OpenAI’s GPT-5.6 Pro whether an AI-written essay might ever be related to certainty, and I recommended linking the issue to Turing’s replica video game and to the “reverse” Turing test postured by AI-text detection. GPT-5.6 Pro linked the concern to binary hypothesis screening and analytical choice theory, proposed early structures and titles, determined prospect literature, and created parts of early drafts. I selected the issue, journal, audience, and ultimate instructions of the argument. I chose which examples and outcomes ought to stay main, consisting of the choice to maintain the Liang research study while challenging it with crucial counterevidence. I examined the pointed out literature, turned down or remedied claims that I evaluated unreliable, remedied mathematical and conceptual mistakes, and restructured the argument.

In modifying the manuscript, I reworded most areas and included considerable descriptions. I periodically asked GPT-5.6 Sol to offer language-editing help for areas I had actually reworded. I developed the figures and made the decisions about the mathematics, analysis, sources, company, and phrasing. I take complete obligation for the resulting manuscript.

Before sending the manuscript for publication, I likewise asked GPT-5.6 Pro and Anthropic Claude Fable 5.1 to evaluate it seriously and determine concerns that customers may raise. I examined the concerns they recognized and made corrections where I evaluated them called for.

Regardless of all this support, Schnell withstands the concept that AI is a “co-author.” Rather, he composes, “Calling ‘coauthored’ would present a various confusion. Mathematical tools have actually long made substantive contributions without therefore ending up being authors … The point is not that the software application contributed absolutely nothing. It is that contribution and authorship are various functions.”