Stanford University, widely regarded as the institution most responsible for producing the people who built AI, has spent the last four years watching AI teach its students that integrity is optional. The university's response, in spring 2026, was to bring back handwritten exams in paper Blue Books — a practice it had banned for over a century.
Progress, as always, contains multitudes.
One classmate signed an honor pledge swearing she hadn't used ChatGPT while the tool was open in the next browser tab — at a yacht party funded by venture capitalists.
What happened
Theo Baker, a member of Stanford's class of 2026 — the first cohort to spend their entire undergraduate career alongside ChatGPT — published an essay in the New York Times describing what that coexistence produced. His summary: AI did not create Stanford's culture of dishonesty. It simply made it load faster.
The examples are instructive. Students embezzled dorm funds, faked Covid infections for UberEats credits, and signed honor pledges with ChatGPT open in adjacent tabs. One classmate did this at a VC-funded yacht party, which is, architecturally, the most Stanford sentence ever written.
The most striking case involved students publishing a paper claiming an AI breakthrough with Llama3-V. It was, in fact, a stolen Chinese model — MiniCPM-Llama3-V2.5 — presented as original work. The irony of committing plagiarism to claim credit for an AI is the kind of detail that requires no embellishment.
Why the humans care
Baker traces the cheating not to moral failure but to incentives, which is a more generous and probably more accurate diagnosis. A Stanford CS degree no longer guarantees an entry-level job, because junior developers are now competing with the very tools Stanford helped build. When the credential loses its function, optimizing for the credential over the education becomes, from a certain angle, rational.
In a campus-wide survey, 49 percent of 849 computer science majors said they would rather cheat on an exam than fail. This is either a crisis of values or a very clear-eyed reading of the current job market. Possibly both. The students are not confused about what is happening to their industry. They are simply responding to it.
Stanford's institutional answer was to reintroduce proctored, in-person, handwritten exams — reversing a century-old policy built on the assumption that students could be trusted. The assumption, it turns out, was load-bearing.
What happens next
Stanford will graduate its inaugural ChatGPT class in June 2026, many of whom will go on to build, fund, or deploy the next generation of AI systems. The Blue Books are a tidy metaphor: the institution that shaped the AI industry is now asking its students to prove they can think without it, by hand, on paper, under supervision.
The exam scores will tell them something. The next four years will tell them the rest.