Brown professor says AI cheating likely drove huge test score swing

In short: A Brown University professor says many students likely used AI chatbots to cheat on a take home exam, after scores dropped sharply on an in person final.

What's going on

Roberto Serrano, an economics professor at Brown University, says he suspects widespread cheating with generative AI, tools like ChatGPT that can write answers in a human sounding way (like a fast homework helper that can also make things up).

The situation centers on ECON 1170, a course Serrano describes as difficult. After a deadly shooting on Brown’s campus in December 2025, Serrano decided to use take home exams for the spring 2026 midterm and final. Enrollment jumped to 86 students, and the midterm results were unusually high, with an average score of 96 out of 100 and 40 perfect scores.

Serrano told Inside Higher Ed that past midterm averages were typically between 65 and 80 percent, and he said this take home exam was harder than earlier ones. He also said many answers felt “off” in style, and when he and graduate students tested the prompts in ChatGPT, they saw similar looking responses.

Serrano then switched the final exam to in person. Eighteen students dropped the course and nine did not show up for the final. According to El País, 22 of those 27 students had scored 100 on the midterm. Among students who took the final, the average score fell to 48.

What to watch

Universities are still figuring out where to draw the line between acceptable help and cheating. Brown recently published a report noting that many students use these tools often, but also worry they may hurt learning and thinking skills. Serrano says schools need to respond more strongly, because he believes normalizing cheating leads to long term harm.

Source: Arstechnica