How AI cheating concerns sent Brown exam scores plunging

Brown University economics professor Roberto Serrano moved a final exam in person after unusually high take-home midterm scores raised AI cheating concerns. The class average fell from 96 to 48, intensifying questions about generative AI, assessment, and whether students are actually learning.

WTF Index IDIOCRACY
◄ Terminator 0 Idiocracy 4 ►

The story centers on generative AI undermining academic assessment and raising doubts about whether students are actually learning.

How AI cheating concerns sent Brown exam scores plunging

A Brown University economics class has become a sharp case study in the academic risks of generative AI. After unusually strong results on a take-home midterm, professor Roberto Serrano changed the final exam to an in-person format. The result was a dramatic collapse in scores.

The episode is not just about one class. It points to a larger problem universities now face: when AI tools can produce plausible answers outside the classroom, grades may no longer show whether students understand the material.

Why the midterm raised alarms

The class was ECON 1170, described in the source article as a quite difficult course. Serrano had allowed take-home exams for both the midterm and the final in spring 2026 after a gunman attacked Brown’s campus in December 2025 and killed two people, including one person who had recently introduced herself to him.

That change in assessment coincided with a striking enrollment shift. The course had never had more than 30 students enrolled at a time, and in some cases had only eight. This semester, 86 students signed up.

Then came the midterm. It was administered on March 5, and the average score was 96 out of 100. Forty students scored a perfect 100.

Those numbers were far outside the pattern Serrano expected. He told Inside Higher Ed, “Historically the average grade in the midterm of this course has ranged between 65 and 80 [percent], and this exam was harder than the exams I wrote in the past, because… take-home is an opportunity to challenge the class a little bit more, given that you’re giving the students unlimited time.”

The written answers also raised concerns. Serrano said many had a “very convoluted style.” When he and his grad students put the exam questions into ChatGPT, they received similar results.

The final exam changed the evidence

Serrano did not immediately void the midterm. Instead, he told students he would use the final exam as a test of whether the midterm distribution reflected real mastery.

In an email to the class, he wrote, “I am not declaring [the midterm] void for now. I am going to give the class a chance to prove me wrong. That is, if the distribution of the final exam is roughly similar to the distribution of the midterm, I will count the midterm. Otherwise, which is of course what I expect to happen, I will declare the midterm void and reweigh the final accordingly.”

After the final was moved in person, 18 students dropped the course. Nine others did not attend the final exam. Of those 27 students, El País noted, “22 had scored a perfect 100 in the midterm exam.”

Among the students who did take the final, the average score fell from 96 to 48. That gap is the central fact making the Brown case so difficult to dismiss. The same course, with the same students, produced radically different outcomes once the exam environment changed.

What the Brown case says about generative AI in class

The source article also points to a broader pattern. A recent survey of Princeton students found that 29.9 percent admitted to cheating with AI on at least one exam or assignment. Brown’s case adds a more detailed view of how suspected AI cheating can appear inside a specific course.

The issue is not that students are incapable of learning. The problem is that generative AI can make a shortcut look efficient, especially for students who are competitive, ambitious, and overscheduled. If a take-home exam can be completed with outside AI assistance, the score may reward tool use instead of understanding.

For instructors, that creates a difficult assessment problem. Take-home exams can allow deeper and more challenging questions because students have unlimited time. But the same format can make it harder to know whether the submitted work reflects the student’s own reasoning.

The Brown episode shows several warning signs universities may need to examine carefully:

  • Sudden score jumps that break sharply from the course’s historical range.
  • Large numbers of perfect scores on a difficult exam.
  • Answer styles that appear correct but unusually convoluted.
  • Major performance drops when students must complete work in person.
  • Student withdrawals or absences after assessment conditions change.

None of those signs alone proves every individual case. Together, in Serrano’s class, they created a picture that he considered serious enough to keep pressing publicly.

Students also worry about the effect on learning

Brown is already examining generative AI in teaching and learning. The source article says the university recently released a provost-led report on “Generative AI in Teaching and Learning,” and that the report found concerns among students as well as professors.

The report noted that “56 percent of undergraduate respondents [at Brown] and 67 percent of graduate and medical student respondents reported intentionally using GenAI tools daily or weekly.” At the same time, large majorities of students had “concerns about the impact of GenAI use on their learning” and a “fear of negative consequences for their cognitive capacity.”

That tension matters. Students may use generative AI regularly while still worrying that it weakens the very skills a university is supposed to develop. The Brown case gives that worry a concrete academic form: if AI helps produce a top score on a take-home exam, the grade may hide a lack of preparation for unaided work.

Serrano’s warning is about more than grades

Serrano, who was born in Spain, went blind from retinal dystrophy at age 17. According to El País, he learned Braille and his academic record opened the doors of Harvard. He described his disease as a restriction to optimize around, not something to over-dramatize.

That background helps explain the force of his reaction. For Serrano, the issue is not simply classroom discipline or grading fairness. It is whether elite education still asks students to think for themselves when shortcuts are available.

He told Inside Higher Ed, “We cannot afford to have a society in which a significant fraction of our best young minds think that cheating is okay,” adding, “That leads to a declining society, to a failed society.”

His final warning was even shorter: “We cannot choose to become idiots.”

Universities now have to decide how to respond to that challenge. The Brown case suggests that generative AI has made exam design, honor codes, and learning outcomes inseparable. If assessments cannot distinguish between student reasoning and AI output, then grades become less trustworthy and learning becomes harder to verify.