Dartmouth Provost Resigns After AI Text Review

On October 30, 2026, Dartmouth College provost Santiago Schnell resigned. The reason was an internal review of his scholarly and public texts. A committee of faculty members established that Schnell substantially reworked materials with AI, did not inform publishers about this, and downplayed the extent of use. This also applies to a column about AI plagiarism published in the Washington Post on August 9, 2026.

The difference from previous cases is that the review was not conducted with AI detectors. The committee's official report specifically states that it did not use detectors. Instead, faculty members examined the query history in the corporate Claude account, correspondence, and other materials. Previously, Semafor checked the column with the Pangram service — the detector rated the text as entirely generated by a neural network. The Dartmouth obtained similar results for other publications by the provost. Schnell objected: detectors are unreliable. He acknowledged using ChatGPT but claimed that he had only refined a few arguments and checked grammar. The committee relied on documentary traces rather than probabilistic assessments.

In practice, this means that internal reviews can be built on logs and correspondence rather than on detector conclusions. For authors, this is a signal: if a corporate account stores query history, it can become evidence. For publishers, it means that declarations about AI use are verifiable. For universities, it means that rules for disclosing AI use apply not only to students but also to administration. Detectors remain a disputed tool: even the committee's report excludes their use.

What remains unknown: the full scope of the committee's report has not been published, nor have the exact wording of the publisher rules that Schnell violated. It is unclear whether the query history and correspondence examined by the committee will be published.