The Hidden Cost of Disclosure: A Multi-institutional Study on Undergraduate Students' Generative AI Usage and Faculty Accusations

Synthesis: Transparency about GenAI use carries a hidden cost: students who disclose may be more likely to be accused. Vetter et al. (2026), a multi-institutional survey of 560 undergraduates across four Northeastern U.S. institutions, find that nearly two-thirds of students never or rarely disclose GenAI use to instructors, and that punitive or ambiguous course policies encourage concealment. Strikingly, students who "always" disclosed their AI use had over three times the odds of being accused by their instructors — an ironic pattern where greater transparency correlates with greater suspicion, revealing a fragile Trust relationship and the hidden costs students incur for being honest.

Key Findings

• Most students do not disclose. 66% of GenAI-using students never or rarely disclosed their use to instructors; only ~10% always did. Frequent (daily/weekly) users were significantly more likely to rarely disclose and less likely to always disclose.
• Punitive and unclear policies drive concealment. Students were more likely to never disclose when they were "not aware of course policies" or when AI was "not allowed." A "limited use in certain situations" policy was the strongest positive predictor of disclosure. Nearly 30% were unaware of their courses' AI policies or had policies unaddressed in the syllabus.
• Transparency is risky. Students who always disclosed had over 3× the odds of being accused. Moderate users were most likely to be wrongly accused; heavy and always-disclosing students drew suspicion. Fear of academic penalty was the most common reason for non-disclosure.
• Students want clarity and collaboration. Most students want clear, unified policies that allow GenAI for brainstorming/research/studying, and support collaborative policy development between faculty and students.
• Trust is fragile. While 92% of students viewed AI as at least somewhat positive for learning, the fear of retribution and inconsistent policies chill honest disclosure.

What this means for practice

• Instructors. Publish a clear AI policy in the syllabus and revisit it aloud: nearly 30% of students were unaware of their course's policy or found it unaddressed, and not knowing the policy predicted non-disclosure.
• Instructors. Replace detection with trust-building conversation — students who always disclosed had over three times the odds of being accused, so disclosure must never be treated as a reason for suspicion.
• Faculty developers. Train staff to reframe disclosure as collaborative articulation of how and why GenAI was used, positioning it as part of learning rather than a confession.
• Administrators. Co-develop unified policies with students: a "limited use in certain situations" policy was the strongest positive predictor of disclosure, whereas punitive or vague policies motivated concealment.
• Administrators. Audit accusation patterns for equity, because moderate users were the most likely to be wrongly accused and the most transparent students drew disproportionate scrutiny.

Limitations

• The survey covers 560 undergraduates at four institutions in the Northeastern U.S.; the authors state this institutional profile limits generalizability to other regions and types of institution.
• The study was advertised as a GenAI survey, so volunteers were likely those already familiar with ChatGPT and Gemini — students anxious about or inexperienced with the tools probably under-responded, which likely inflates the reported non-use rate.
• All variables are self-report on sensitive, misconduct-adjacent behavior: despite anonymity assurances, some respondents may not have fully reported their GenAI use, and gift-card incentives may have encouraged fast completion.
• The disclosure–accusation findings are observational and statistically weak (Cramer's V = .13–.22 across the reported chi-square tests), so they establish association rather than a causal effect of transparency on accusation.

Citation

Vetter, M. A., Farag, I., Jiang, J., Lucia, B., & Silvestro, J. J. (2026). The Hidden Cost of Disclosure: A Multi-institutional Study on Undergraduate Students' Generative AI Usage and Faculty Accusations. SSRN.