AI Pulse by Inblix

Brown prof's AI cheating test: grades crash from 96% to 48%

The Decoder · Jul 12, 2026 · 2 min read · Read original article →

Curated by the Inblix editorial team


Featured image for article: Brown prof's AI cheating test: grades crash from 96% to 48%

When Roberto Serrano’s take-home midterm averaged 96 percent, he knew something was off. The Brown University economics professor had been teaching the course for years, and scores typically hover between 65 and 80 percent. So he did what any suspicious instructor would do: he fed the questions into ChatGPT. The AI’s answers matched his students’ work almost exactly, right down to a strangely convoluted mathematical proof that most humans wouldn’t choose.

The real reckoning came with the final. Serrano made it a proctored, in-person exam. Eighteen of his 86 students dropped the course before test day, and nine others simply didn’t show up. The class average crashed to 48.6 percent — the lowest score he’s ever recorded. Nineteen students failed outright. Serrano voided the midterm entirely and made the final worth 80 percent of the grade. Brown’s administration told him to report each cheating case individually, a response he called “meek” and “ridiculous.”

This isn’t one professor’s anecdote. Two large-scale studies back up the pattern with hard numbers. A Chinese study tracking over 26,000 students in grades 7 through 12 found that six months after students gained access to AI tools, homework scores shot up 18 percent while completion time dropped from 64 to 45 minutes. Exam scores, meanwhile, fell by 20 percent. On high-stakes entrance exams, the hit was even worse — an 18 to 24 percent decline that took about two years to fully materialize. Top students suffered most, losing nearly a quarter of their performance.

A UC Berkeley study examining more than 500,000 grades at a Texas research university tells a similar story. After ChatGPT launched, courses heavy on writing and programming assignments saw a 13 percentage-point jump in A grades. When researchers compared homework-heavy courses to exam-heavy ones, the gap hit 16 percentage points. The message is unambiguous: unsupervised assignments are now an AI free-for-all. Serrano put it bluntly to Inside Higher Ed: “We cannot afford to have a society in which a significant fraction of our best young minds think that cheating is OK. That leads to a declining society, to a failed society.”

💡 Key Takeaways

  1. When a Brown economics professor switched from take-home to proctored exams, the class average plummeted from 96 percent to 48.6 percent — and 27 of 86 students either dropped or no-showed.
  2. A Chinese study of 26,000 students found AI use inflated homework scores by 18 percent while exam performance dropped 20 percent, with the worst effects hitting top students.
  3. UC Berkeley research on 500,000 grades shows ChatGPT caused a 13 percentage-point surge in A's for writing-heavy courses, concentrated entirely in unsupervised assignments.
  4. Brown University's insistence on case-by-case cheating reports signals that institutional responses to AI dishonesty remain reactive and toothless.

Keep reading: See related articles below for more coverage on this topic.

Get smarter about AI

The sharpest AI news, curated daily. Delivered free to your inbox.

Learn more

Glossary terms

← Back to all articles