AI Shortcuts and Math Gaps Drive Record Failure Rates in Berkeley CS

UC Berkeley computer science courses saw a record-breaking spike in failing grades during the spring 2026 semester, far exceeding historical department averages. Faculty blame this trend on a combination of AI-driven cheating, a decline in foundational math skills, and reduced student support due to understaffing. To combat these issues, professors are calling for the return of standardized testing and urging students to re-engage with difficult problem-solving without automated assistance.

Key Points

• Failing grades in introductory and upper-division CS courses have soared far beyond the department's 5-7% guidelines, with some classes seeing failure rates triple.
• Professors attribute the decline to the 'vast increase' in AI-related academic dishonesty and students using LLMs as a crutch rather than learning the material.
• Faculty report a significant lack of mathematical preparedness in students, leading to a systemwide petition to reinstate SAT/ACT requirements for STEM admissions.
• Understaffing due to high TA wages has led to the removal of high-scoring projects and reduced instructional support.
• Student engagement has plummeted, with professors reporting empty office hours despite the high rate of academic failure.

Sentiment

The overall sentiment leans toward agreement with the article's concern, but not with a single-cause explanation. Commenters are broadly worried about AI-enabled deskilling and weak technical preparation, while also arguing that Berkeley's failures likely reflect several overlapping institutional and educational problems. The mood is pessimistic and contentious, especially around admissions and standardized testing, but there is a constructive thread around redesigning assessment so students must prove real understanding.

In Agreement

• LLMs make it easy for students to complete assignments without developing the reasoning and debugging skills that exams require.
• AI dependency may be eroding independent thought beyond school, with professionals also leaning on models for writing, coding, brainstorming, and communication.
• Take-home programming and math work are increasingly weak assessments unless courses require students to show process and defend their understanding.
• The article's concern about weaker math preparation resonates with commenters who see students arriving in technical courses without enough prerequisite problem-solving practice.
• Universities should respond with more supervised assessment, direct explanation, and support for mastery rather than assuming polished submissions reflect learning.
• TA shortages, large classes, and credential-focused incentives make it easier for students to fall through the cracks and harder for instructors to catch shallow understanding.

Opposed

• AI may be only a convenient scapegoat for deeper problems such as pandemic disruption, weak K-12 preparation, grade inflation, admissions policy, and student motivation.
• Cheating and shortcut-seeking are not new, and widespread AI use may reflect changed norms around tools rather than a sudden collapse in honesty.
• LLMs can be excellent tutors when used to explain concepts, provide examples, or give feedback without replacing the learner's own effort.
• Coursework and assessment may need to adapt to advanced tools instead of treating all AI use as illegitimate.
• Some commenters argue that failure should be treated as a useful educational signal and handled compassionately rather than framed only as moral decline.
• Politicized claims about testing, equity, and admissions are contested, with some commenters warning that they oversimplify a complex preparation problem.