AI detection tools are pushing a fierce question into universities, newspapers, and publishing: who really wrote the words on the page? Two separate controversies involving Dartmouth College provost Santiago Schnell and Canadian-Haitian author Thelyson Orelien show how explosive that question becomes when detection scores collide with human denials.
In Schnell’s case, Pangram identified an article published in The Washington Post as 100 percent AI-written. The finding followed an analysis of his published work, with a clear dividing line around November 2022, when ChatGPT was publicly released.
Santiago Schnell Faces Questions Over Published Work
All of Schnell’s works written before November 2022 were found to be 100 percent human written. Nine works produced after that date received a median AI-written score of 96 percent, creating a major gap between his earlier and later writing.
Schnell admitted using chatbots in his writing process, but described AI’s influence as modest. He said he used the tools for language refinement and copyediting rather than for developing the central ideas or research behind his work.
“I develop ideas and arguments, conduct the research, evaluate the evidence and sources, determine the structure, prepare drafts and decide what appears in the final text,” Schnell said.
He also stated, “I regard these technologies as assistive tools,” and added, “I am fully responsible for any text under my signature.” Schnell later wrote a follow-up explaining that he first prepared a draft, then used an AI tool to assist with language and clarity.
That explanation did not settle the controversy. Schnell’s statement itself was identified as 100 percent AI-written, adding another layer to the dispute over how much assistance he received and whether his description matched the detected text.
Students and critics called for Schnell to be fired, describing his AI usage as an insult to the Dartmouth community. The issue carries extra weight because Dartmouth students would violate the Academic Honor Principle if they used AI to write papers without acknowledgment.
Orelien’s Novel Raises a Different Detection Battle
The controversy surrounding Thelyson Orelien centers on his novel and the claim that Pangram concluded the work was largely AI-generated. Orelien denies using AI tools to write the novel and plans to prove that he wrote it himself.
Excerpts from the book were tested with various AI detection tools, producing mixed results that ranged from “very likely” to “100 percent” AI-generated. Benoit Raphael, an entrepreneur and writer specialized in AI, tested Orelien’s work with Pangram and found results indicating a high likelihood of AI authorship.
Critics also pointed to the book’s repetitive sentence structure and overuse of analogies as signs of AI authorship. Fabrice Colin reviewed Orelien’s work, while podcast host Lea Bory commented on the writing, adding public scrutiny beyond the automated scores.
Orelien rejects that reading of his process. “I didn’t write this novel with my head, but with my guts and my heart,” he said, turning the dispute into a direct challenge over whether software can recognize a writer’s personal voice.
What Can AI Detectors Really Prove?
AI detection tools compare large collections of AI-generated and human-written texts to identify patterns associated with machine-produced writing. Pangram says it examines subtle cues, including linguistic tics, repeated words, and monotonous sentence structures.
Thierry Poibeau, a researcher at CNRS and an NLP specialist, explained that detection systems look for patterns such as triadic phrasing and em dashes, which were common in AI-generated texts. Those signals can help investigators identify writing that shares features with generated text, but they do not automatically establish who wrote it.
The systems face a moving target because AI programs evolve and can mimic a writer’s style. Many accuracy figures published by AI detection tools cannot be verified, and they do not account for false positives, when human writing receives an AI label.
A study published by the University of Chicago in October 2025 offered another piece of the puzzle. Pangram achieved near-zero error rates on medium-to-long extracts in that study, a result that strengthens its case for certain types of analysis without turning a detection score into a final verdict.
Max Spero, CEO and co-founder of Pangram, said the tool should serve as a starting point for investigation, not the final judgment. That distinction matters in both controversies: a score can trigger questions, but people still need to examine drafts, editing steps, evidence, and the writer’s account.
The Schnell and Orelien disputes point toward a future where authors, academics, editors, and students must explain how AI tools shaped their work. Detection technology will keep advancing, but so will the need for clear rules and honest disclosure. The central challenge is no longer just spotting AI writing; it is deciding what kind of assistance a community accepts, and who remains responsible for the final words.
Based on
- Provost of Dartmouth Busted Using AI for His Own Writing in Newspapers and Academic Journals — futurism.com
- Spotting AI writing: how reliable are the detectors? – The Japan Times — japantimes.co.jp
- Thelyson Orelien: How reliable are AI writing detectors? — ctvnews.ca
- Spotting AI writing: how reliable are the detectors? — france24.com
