Several UC Berkeley computer science courses recorded unusually high failure rates in spring 2026, prompting instructors to examine the effects of generative AI use, mathematical readiness, staffing limits and declining participation outside class. The figures departed sharply from recent results and the department's typical grading guidance.
Data reported by Berkeleytime showed that 35.3% of students in CS 10 received F grades, as did 10.6% of students in CS 61A. In spring 2024 and spring 2025, neither course had an F rate above 10%. Both 2026 classes averaged a C-plus, equivalent to a 2.3 grade-point average. Department guidance says lower-division courses would typically award D or F grades to 7% of students and have an average GPA between 2.8 and 3.3.
Teaching professor Dan Garcia, who taught both courses, said he believed a major cause was increased academic dishonesty involving large language models. Nearly 30 CS 10 students were caught cheating on take-home exams, he said, while others appeared to have depended on AI for coursework and then lacked the same support during exams. The classes used published point thresholds rather than curves, meaning poor performance was not offset by classmates' results.
The pattern extended beyond introductory programming. EECS 127, an upper-division optimization course taught by associate teaching professor Gireeja Ranade, posted a 16.8% F rate. Department guidance describes a combined D-and-F rate of about 5% as typical for upper-division classes. Ranade said students struggled with prerequisites including linear algebra, vector calculus and mathematical proofs. The campus mathematics department said it had no record of the open-internet, open-AI assessment policy described to Ranade by one student; the newspaper could not verify whether a similar policy existed elsewhere.
Staffing also changed the course environment. Ranade removed a final project after losing teaching-assistant support, eliminating an assignment on which students had generally scored well. Both instructors reported sparse office-hour attendance despite invitations to participate.
Garcia plans to explain the spring results to future classes and seek ways to identify students needing remedial help. Ranade argued that the response to AI should be more teaching and stronger analytical preparation, not reduced expectations. Their accounts do not establish a single cause for the grade shift, but they identify overlapping academic, technological and institutional pressures that the department will have to disentangle.\n\nThe published percentages also require careful interpretation. They describe outcomes in three named courses during one semester, not the performance of Berkeley students generally. Likewise, instructors offered several possible contributors rather than a controlled analysis isolating AI use. Future semesters, conduct-case outcomes and changes in staffing or assessment design may help show which explanations account for the largest share of the shift.
