PrincetonOne faculty member opposed Monday’s proctoring vote.
AI tools made cheating invisible to peers; a 2025 senior survey found 30% had cheated, leaving the peer-reporting system ineffective.
Proctors observe from July 1 but may not intervene; suspected violations go to the student-run Honor Committee.
Sources: Hacker News
THE ORIGINAL BARGAIN
Princeton's 1893 honor code traded faculty surveillance for student self-policing: students sign a pledge and are obligated to report peers. The system survived 133 years because cheating left visible traces — a wandering eye, a passed note, a whispered answer.
WHY PEER REPORTING WORKED
Honor codes function as equilibrium games: they hold when most students believe most others comply. Once perceived cheating crosses a threshold, the cooperative norm collapses — non-cheaters feel like suckers and either start cheating or stop reporting. Game theorists call this a tipping model.
WHAT AI BROKE
Earlier cheating — crib sheets, copied answers, even contract essay mills — produced artifacts another student could see. A classmate using ChatGPT on a laptop or phone under the desk produces no observable signal. The peer-detection layer the honor code depended on no longer has anything to detect.
THE PEER GROUP
Among elite US universities, true unproctored exams were always rare. Caltech and the US service academies maintain honor codes with broad take-home privileges; UVA's single-sanction code (expulsion for any violation) is the strictest; most of the Ivy League proctors normally.
THE DETECTION ARMS RACE
AI-detection tools (Turnitin, GPTZero) report false-positive rates that make them legally unusable for discipline — a single wrongful expulsion invites litigation. Watermarking schemes proposed by OpenAI and Google have not shipped at scale. There is currently no reliable forensic method to prove a given essay was AI-generated, which is why universities are retreating to the one thing that works: watching students take the test in person.
THE DEEPER SHIFT
The proctoring vote is a small instance of a wider pattern: institutions that ran on trust-based verification — open-book exams, take-home finals, unsupervised remote work — are rebuilding surveillance layers they had spent decades dismantling. The cost is not just supervision but the signal that supervision sends about what students are presumed to be.