The Witness in the Room

On Monday, May 11, Princeton's faculty voted, with one opposing voice, to end 133 years of unproctored examinations. Starting July 1, all in-person exams at the university will be supervised by instructional staff present in the room.¹ The Honor Code, introduced in 1893, did not require this. It asked only that a student sign a pledge.

The faculty did not name a single cause, but the conversation on campus and in the coverage pointed in one direction. The Daily Princetonian's 2025 senior survey, of 501 graduating seniors, found that nearly 30% reported cheating on an assignment or exam during their time at Princeton. Twenty-eight percent reported using ChatGPT on an assignment that specifically prohibited it, more than double the figure from the year before.²

The quieter pair of numbers from the same survey is the heart of the story. Forty-five percent of seniors said they knew of Honor Code violations they had chosen not to report. Only 0.4% had ever reported another student.²

Long before AI was the breaking point, the system was running on a polite agreement. Everyone signed the pledge. Almost no one carried a report to the committee. The Code held because nobody was asked to look too closely. Generative AI did not invent dishonesty in the room. It made dishonesty so cheap, so plausible, and so quietly undetectable that the agreement of not-noticing could no longer hold.

What the Pledge Was Always Verifying

A handwritten exam booklet is a tiny artifact. It captures forty minutes of one Tuesday afternoon. The Honor Code asked us to treat that artifact as if learning lived inside it, as if the test were proof of the thinking behind it. For 133 years that was sufficient, not because the artifact was rich, but because we could not see anything else.

Now we can. AI, almost as a side effect, has revealed how thin the artifact always was. A student can submit a perfect essay produced in three minutes by a model in another tab. Another student can submit a careful, struggling, ten-draft essay produced over four days. The two final documents may look identical. The Honor Code asks the same question of both. The question can no longer answer itself.

Princeton's response is the most conservative one available. It puts a body back in the room. A proctor watches the moment. The test result is verified by the presence of a witness, not by the integrity of the artifact. It is an honest admission that the artifact alone has stopped speaking, and it solves the problem by reaching backward, toward a pre-AI version of the same thin proxy.

The Other Direction

Across the country, blue books are coming back. The University of Virginia bookstore reported a roughly 27% jump in blue book sales between April 2025 and April 2026, with Texas A&M, the University of Florida, and UC Berkeley reporting the same trend.³ The instinct is clear. If we cannot trust the artifact, we will return to the artifact we can at least watch being produced.

There is another direction, and it is the one quietly forming inside classrooms that have lived with AI long enough to know what it is good at and what it is not. The moment of production was never the meaningful unit of learning. It was simply the unit we could record. What if the unit we could record were the whole act of learning instead.

This is what visible work makes possible. When a student writes inside a tool that captures their drafts, their pauses, their backspaces, their questions to a tutor, their corrections after feedback, their second thoughts in the margins, the question of "did the student write this" recedes. A richer question moves in front of you. The journey is the artifact. The witness is not a body in the room. It is the record of the work itself.

This is the bet we have made with Koan. A teacher can read back what a student asked Aidan, where they paused, where they pushed back, where they revised. The record is not surveillance. It is the slow, patient accumulation of learning, in light, where it can be seen.

Princeton chose to return to 1893. The path forward, for most schools, is to keep walking. Not back to a witness in the room, but toward a record of the work that does not need one.

If a student's whole journey to a final answer were visible, would we still need to ask whether the answer was theirs?