Why So Many Students Are Failing Computer Science Classes at UC Berkeley

UC Berkeley's computer science program saw a dramatic increase in failing grades during spring 2026. About 35% of students in CS 10 received F grades, and 11% failed CS 61A. According to The Daily Californian, these numbers are much higher than the previous two years, when failure rates stayed below 10%.

The department has guidelines that say roughly 7% of students should get D's and F's combined. But both courses ended up with average grades around C-minus. The department was aiming for a C-plus range.

The Problem Spreads Beyond Intro Classes

This isn't just happening in beginner courses. An upper-level optimization class called EECS 127 saw about 17% of students fail. That's more than three times what the department normally expects.

Dan Garcia, a professor who taught two of these struggling classes, reported finding nearly 30 students cheating on take-home exams in CS 10. Garcia created the curriculum these courses are built on.

What's Causing the Drop

Faculty point to two main culprits: students using AI tools like ChatGPT, and students arriving without strong math skills.

The response has been substantial. More than 1,300 UC professors signed a petition raising concerns about these academic problems.

A separate study at Berkeley found that A grades shot up by roughly 30% right after ChatGPT launched, before the university could detect and respond to AI usage. Professors also mention that not having enough teaching staff makes things harder.

It's worth remembering that this kind of disruption has happened before. When calculators became widespread in the 1980s, when Google arrived in the 1990s, and when smartphones spread in the 2000s, each one created a moment where students could suddenly access answers faster than teachers expected. Schools eventually adapted by changing how they tested and taught. But this AI shift seems to be moving faster than those earlier changes did.

What This Means for the Pipeline

These high failure rates matter because CS 10 and CS 61A are gateway courses. CS 10 is for anyone curious about how computers work. CS 61A is the foundation for students planning to major in computer science. If students are failing these courses at much higher rates, fewer people will move forward into CS majors and related fields.

The math problem may be deeper than just AI. Students in recent years have relied more on calculators and digital tools to do math, so some arrive at college without the foundational math they need for computer science. Designing algorithms and understanding how fast they run both require solid math skills.

How Classes Test Students Is Changing

AI tools can now write working code from plain English descriptions. That's flipped the whole game for how teachers assess what students know. Homework assignments that students used to do at home, which seemed safe from cheating, now face a new challenge: the computer can just solve it.

Professors face a real puzzle here. AI tools can write code explanations, help debug problems, and even help study for exams. Where's the line between using a tool fairly and cheating? It's getting hard to tell.

The problem isn't just Berkeley's. Universities across the country are struggling with the same issues, which is why so many faculty signed that petition.

The situation right now probably won't last forever. Schools have historically found ways to adapt when new technology changes how students learn and work. The real challenge is whether they can adapt fast enough to keep computer science education strong while still letting people access and learn the subject.

Some colleges are trying new approaches: real-time video monitoring during exams, focusing more on whether students understand ideas rather than whether they can write perfect code, and actually teaching students how to use AI tools the right way rather than just banning them. A few places are even creating separate "AI-allowed" and "no AI" versions of assignments, recognizing that real software work in the future will involve AI.

The numbers from Berkeley give us an early look at how AI is shaking up computer science education. Right now it looks disruptive, but history suggests schools will find new balance points as both students and teachers get used to learning alongside AI tools. The key will be making sure students still build the real problem-solving skills that matter, regardless of what tools they use.