<p>To understand why, you need to understand what the thesis was actually doing. It was never primarily about the text. The text was a proxy—an externally verifiable trace of an internal process: months of reading, arguing with a supervisor, hitting dead ends, restructuring an argument at eleven at night because a seminar colleague had pointed out a fatal flaw. The written product was credible as an assessment instrument only because it was genuinely difficult to fake. That condition no longer holds. A large language model produces, in seconds, prose that is more fluent, better referenced, and more structurally coherent than most submitted dissertations—and it does so at near-zero cognitive cost to the person submitting it. Assessing the resulting document is like judging a painting competition in which half the entrants have quietly swapped in prints from an online gallery: the judges are evaluating something real, but not what they think they are evaluating.</p>