Dartmouth University vice dean resigns over AI use

Summary

Santiago Schnell, Dartmouth College's assistant provost, was forced to resign in October 2026 after an internal investigation found that he had extensively used generative AI in published articles without disclosing it, as editors had requested. The irony is heightened by the fact that Schnell had previously warned against AI-assisted academic dishonesty in a Washington Post op-ed weeks earlier.

Why it matters

This incident sits at the intersection of academic integrity, AI transparency, and the reliability of AI detection tools. It raises urgent questions about where the line is drawn between legitimate assistance and undisclosed authorship. The situation is particularly striking given Schnell's role as the institution's chief academic administrator, as his actions directly challenged the very policies he was responsible for enforcing. The irony between his public stance against AI plagiarism and his own undisclosed use marks a defining moment for how universities and publishers will handle AI-generated content in the future.

Highlights

• Schnell's Washington Post op-ed warning against AI plagiarism was flagged by Pangram as 100% AI-written. The median AI content score for his nine texts dated 2026 was measured at 96%.
• The investigation committee reviewed Anthropic Claude account logs provided by the university. It was determined that Schnell had explicitly asked Claude whether his drafts contained 'AI writing signatures.'
• Schnell's defense focused on the unreliability of AI detection tools. The article illustrates this general concern by citing the case of a journalist who was falsely flagged by a detection tool.

Critical take

The article relies heavily on Pangram's 96% median probabilistic detection score as the backbone of its 'evidence.' However, it simultaneously acknowledges that these tools lack reliability. This tension undermines the narrative's own certainty. While the sidebar note regarding journalist Kimberly Gasuras is illuminating in demonstrating the flaws of detection tools, it shifts the focus away from the actual institutional investigation and the documented account log findings.