Can students cheat their way to a biology degree? A case study of the vulnerability of biology course grades to academic dishonesty in the era of generative AI

Synthesis: Chan and colleagues coded the syllabi of every offering of one biology department's four core required courses — 38 syllabi in one 2025–2026 term, a 100% collection rate — and surveyed its instructors on which graded categories they consider vulnerable to academic dishonesty. Only in-person proctored exams were perceived as minimally vulnerable; on average about a third of a student's grade (33%) was highly vulnerable and roughly 80% (81%) at least somewhat vulnerable to academic dishonesty. Outside-of-class exams were rated similarly vulnerable whether proctored or not, and the low-stakes assignments that evidence-based assessment reform has promoted for thirty years are what instructors see as most exposed. The authors frame this as a grading-design problem that threatens the validity of course grades, not a student-morality problem.

Context: Two Changes That Compound

Thirty years ago a biology grade rested on a few high-stakes, in-person proctored exams that made dishonesty difficult and risky. Following national calls to replace high-stakes testing with multiple low-stakes formative assessments (AAAS's Vision and Change, Bransford et al., 2000), grades were rebuilt around frequent assessment, and worries about cheating were set aside. The authors call this the pre-GenAI assumption and test whether it survives. Two changes compound it: the pandemic-era normalization of remote delivery, which moved assessment outside the classroom; and ubiquitous generative AI, which makes accurate answers easy to obtain while remaining hard to detect, lowering the perceived risk that earlier deterrents relied on.

Design: Syllabi Plus Instructor Perception

A large research-intensive US institution, whose online and in-person biology degrees are treated as equivalent, requested the study. First, the team collected a syllabus for every offering of all four courses in one 2025–2026 semester (38 syllabi, 100% collection rate) and coded each for graded categories and their point shares, sorted by class setting, proctoring and stakes into five: outside-of-class assignments, in-class assignments, unproctored and proctored outside-of-class exams, and proctored in-person exams. Lab components (15–40% of the grade) were excluded because syllabi did not disclose their breakdown. Second, the chair surveyed every instructor who taught any biology course that year, and 56 responded (47%), each rating the largest course they taught from minimally vulnerable (0) to highly vulnerable (4); differences were tested with a Kruskal–Wallis test (χ²(4) = 67.4, p < .001) with pairwise Wilcoxon tests.

How the Grade Is Weighted

Among the 21 in-person syllabi, all included outside-of-class assignments worth a mean of 26.6% (SD 9.1) and 90.5% included in-class assignments at 14.6% (SD 6.9). In-person exams appeared in 52.4% of courses at 66.5% (SD 14.8), proctored outside-of-class exams in 38.1% at 54.2% (SD 2.4), and unproctored outside-of-class exams in 9.5% at 50% (SD 14.1). Among the 17 online syllabi, every course had outside-of-class assignments worth 40.8% (SD 7.8) and proctored outside-of-class exams worth 59.2% (SD 7.8), and none used unproctored outside-of-class exams. More than half the syllabi (61%) came from courses enrolling 200 or more students and 45% were fully online, so the aggregate is weighted toward the largest online and in-person offerings.

What Instructors Consider Vulnerable

Ratings were pooled across modalities. Outside-of-class assignments were median 4 (highly vulnerable), proctored and unproctored outside-of-class exams both median 3, in-class assignments median 2 with the widest disagreement, and in-person exams median 0 (minimally vulnerable), the only category seen as essentially protected. In-person exams were significantly less vulnerable than every other category (p_adj < .01), and outside-of-class assignments significantly more vulnerable than in-class assignments (p_adj < .01); instructors read this as in-person proctoring's relative effectiveness rather than immunity. Proctored and unproctored outside-of-class exams were rated equally vulnerable, showing no perceived benefit from lockdown browsers. Prior literature is mixed, and those tools carry documented costs: anxiety, privacy concerns, technical problems and false accusations.

How Much of a Grade Is Exposed

Combining weights with ratings: conservatively, treating only the highly vulnerable category, an average of 33% of course points were highly vulnerable: 27% in person and 41% online. Less conservatively, counting the four categories rated more vulnerable than in-person exams, an average of 81% of course points were at least somewhat vulnerable — about 65% in person and 100% online. This is not a claim that students are cheating, only that this share of the grade could be completed dishonestly. Because instructors retain autonomy over grading, offerings differ widely: one in-person section had 10% of its points highly vulnerable, another 33%. A student earning a C on exams while cheating to an A on all highly vulnerable work would finish at a C in the first but a B in the second.

Key Findings

1. Only in-person proctored exams were perceived as minimally vulnerable (median 0); every other category was at least somewhat vulnerable (χ²(4) = 67.4, p < .001).
2. Outside-of-class assignments were perceived as highly vulnerable (median 4) and significantly more vulnerable than in-class assignments (p_adj < .01).
3. Proctoring outside class made no perceived difference: proctored and unproctored outside-of-class exams both rated median 3.
4. A third of the grade was highly vulnerable and 81% at least somewhat vulnerable — 27% versus 41% highly vulnerable in person versus online.
5. Identical performance can earn different grades in different sections of the same course, because point weights for vulnerable work sit with instructor autonomy.
6. The authors locate the problem in grading structure, not student character, calling for assessment design that preserves grade validity and departmental responsibility.

Recommendations and Their Trade-offs

The proposals are graded by severity, each paired with its cost. First, shift points from vulnerable low-stakes assessments toward work that more reliably reflects learning, such as proctored in-class work — though removing the incentive can remove the reason engaged students did the work. Next, ungrading uncouples assignments from points entirely, which they theorize reduces stress and supports intrinsic motivation, though a pressed student can still do low-stakes work with AI and no real cognitive effort. Eliminating outside-of-class assignments is the clearest trade-off: it removes the incentive for dishonest completion but also for honest effort. Their preferred direction keeps incentives while enhancing proctoring and accountability, easier in person and in smaller classes. Why do instructors keep using categories they call highly vulnerable? They point to structure: inertia, where pedagogical change is not incentivized; lack of time or knowledge, with Multimodal AI, more accurate models and free institutional access both arriving in September 2024; and resistance when reform conflicts with prior best practice, since redesigning away from graded outside-of-class work may mean giving up flipped classrooms and active learning.

What the design does establish

The design does establish a cheap, replicable and diagnostic method: code each syllabus's graded categories by class setting, proctoring and stakes, then multiply by instructor-perceived vulnerability for a per-course exposure figure — no new data collection, and a number a curriculum committee can act on. It travels beyond biology to most of undergraduate STEM, and invites comparison with summative redesign that moves certified weight onto supervised, process-visible work (AI in Particle Physics Education: Research Problems and Foundational Skills) and with the computing-education finding that redesign, not detection, is where the evidence sits (Generative AI in computing education: A systematic review and a framework for responsible integration).

What this means for practice

• Instructors. Compute your course's exposure before redesigning: multiply each graded category's point weight by instructor-rated vulnerability — the procedure behind the 33% and 81% figures in these 38 syllabi.
• Instructors. Stop treating proctoring as the fix for take-home exams: proctored and unproctored outside-of-class exams were rated equally vulnerable (median 3); only in-person proctored exams came in at median 0.
• Assessment designers. Rebalance points away from vulnerable low-stakes work rather than abolishing it, and name the trade-off: removing the incentive also removes the reason engaged students did it.
• Administrators. Treat grade validity as a departmental responsibility — audit every offering's syllabus, and use circulating assistants for in-class accountability where course size (61% enrolling 200 or more) makes monitoring impossible.
• Researchers. Replicate the syllabus-audit-plus-perception method in other departments, disciplines and terms: one department in one term can only predict, not demonstrate, how widespread the exposure is.

Limitations

• One department, one institution, one term. As a case study it cannot speak to generalizability, only predict that the problem is widespread.
• Perceptions, not observed dishonesty. The vulnerability figures rest on instructor estimates of exposure, so they measure the size of the opportunity rather than its use.
• Lecture components only. The syllabus analysis excludes lab work that constituted 15-40% of the grade in some courses, so the percentages cover the codable graded portion.
• Weighted toward the largest offerings. The sample skews to large courses (61% enrolling 200+) and online sections (45%), where vulnerability concentrates, so the aggregate describes the largest offerings, not a typical small seminar.

Citation

Chan, B. G., Anderson, E. P., Lu, S., Abdellatif, N., Cooper, K. M., & Brownell, S. E. (2026). Can students cheat their way to a biology degree? A case study of the vulnerability of biology course grades to academic dishonesty in the era of generative AI. OSF preprint.