dailycal.org via Reddit

Berkeley CS F-Rates Triple as AI Cheating Surges

education ai ethics hallucinations ai-education ai-cheating

Key insights

  • UC Berkeley's CS 10 failure rate reached 35.3% in spring 2026, more than tripling from under 10% in prior years.
  • Professor Dan Garcia called AI-driven academic dishonesty the 'primary driver' of the spike, with nearly 30 CS 10 students caught cheating on take-home exams.
  • Both EECS professors now support reinstating ACT/SAT testing for STEM admissions, citing deepening foundational math gaps in incoming students.

Why this matters

Berkeley's EECS data gives AI practitioners the first rigorous, named-faculty documentation of LLM use collapsing course pass rates at a top-tier CS program, a concrete signal that AI-assisted work is producing credential holders who may lack the foundational skills the credential implies. For founders and technical leaders hiring from elite CS pipelines, this raises a direct question about what a Berkeley CS degree will signal about a candidate's actual capabilities within the next few graduating cohorts. The dual failure, where cheating inflates apparent learning while foundational math gaps compound in prerequisite courses, means the problem grows invisibly until it surfaces in production systems and on-the-job engineering performance.

Summary

Failing grades in UC Berkeley's CS department hit historic highs in spring 2026, with 35.3% of CS 10 students receiving F's, up from under 10% in prior years. CS 61A reached 10.6% and EECS 127 hit 16.8%, all well above the department's 7% guideline for combined D's and F's in lower-division courses. Professor Dan Garcia named a 'vast increase in academic dishonesty' driven by large language model use as the 'primary driver' of failures, with nearly 30 CS 10 students caught cheating on take-home exams. Associate teaching professor Gireeja Ranade identified a parallel problem: incoming students arriving with foundational gaps in linear algebra, vector calculus, and proofs. Essentially: (Dan Garcia, Gireeja Ranade, UC Berkeley EECS) are contending with AI undermining both academic integrity and baseline student preparation at the same time. - CS 10 failure rate jumped from under 10% to 35.3% in a single semester. - EECS cut undergraduate TA positions due to high hourly wages, forcing Ranade to eliminate a final project component that had historically boosted scores. - Both professors now support reinstating ACT/SAT requirements for STEM admissions. Berkeley's EECS department is a live case study in how generative AI disrupts academic institutions from multiple directions simultaneously.

Potential risks and opportunities

Risks

  • Graduating cohorts who passed with AI assistance but lack foundational linear algebra and vector calculus skills enter engineering roles at major tech employers, a gap invisible until it surfaces in production code and system design.
  • If Berkeley faces political pressure to relax grading standards rather than enforce academic integrity, peer UC campuses and other top programs may follow, eroding the signal value of CS credentials industry-wide.
  • ACT/SAT reinstatement efforts by EECS faculty could face legal or political obstacles under California's ongoing admissions-equity debates, stalling the proposed fix while failure rates remain elevated.

Opportunities

  • Academic integrity detection vendors such as Turnitin, GPTZero, and Copyleaks can use Berkeley's documented failure-rate case to accelerate enterprise contracts with CS and EECS departments at research universities.
  • In-person proctoring and secure exam platform providers gain a clear institutional sales narrative as professors abandon take-home exams in response to LLM cheating at scale.
  • Math-prep and CS-foundations platforms have a concrete, named institutional pain point to address: the linear algebra, vector calculus, and proof-writing gaps that Ranade identified as entering students' core deficiency.

What we don't know yet

  • Which specific large language models were identified in the cheating cases, and whether Berkeley has engaged AI providers to detect or limit exam misuse.
  • Whether the EECS department has a concrete plan to reverse the undergraduate TA staffing reduction, or whether the high-hourly-wage constraint makes it structural and ongoing.
  • Whether comparable failure rate spikes are appearing at peer institutions such as MIT, Stanford, or CMU, or whether Berkeley's numbers reflect local policy changes rather than a sector-wide trend.

Originally reported by dailycal.org

Read the original article →

Original headline: UC Berkeley CS Failing Grades Hit Historic High — 35% of CS 10 Students Got F's in Spring 2026, Professors Name AI Cheating as Primary Driver