It's OK Because...": The Wild West of Student Rationalization of AI Use in Academic Writing

Synthesis: Kim, Toyama, Kim, and Carroll (2026) investigate how students make moral sense of AI use in academic writing through semi-structured interviews (n=20), AI chat logs, and course documents. They identify five distinct "sites" of AI policy — from faculty intention to students' actual practice — where interpretation diverges, and a taxonomy of 23 distinct rationalizations (grouped into six classes) that students use to justify AI use, including conscious violations of course policies. Because these rationalizations are ad hoc and post hoc, modern AI presents a "steep, ethical, slippery slope" that students conceptually slide down, landing far outside instructors' pedagogical goals.

Five Sites of AI "Policy"

For any given course, the paper finds at least five distinct sites that house concepts of appropriate AI use, which often diverge from one another:

1. Faculty Intention — instructors' expectations of student AI use (often not clearly expressed).
2. Formal Policy — the explicit written rules in syllabi or assignment instructions.
3. Student Interpretation — a student's own understanding, recall, and/or interpretation of the formal policy.
4. Student Self-Policy — a student's own normative views about appropriate AI use.
5. Student Practice — a student's actual AI-use behavior.

The gaps between these sites are systemic: faculty intention is not always reflected in formal policy, students misremember or reinterpret policy, and actual practice diverges from both. This reveals why Academic Integrity policies communicated at one site often fail to shape behavior at another.

Taxonomy of 23 Rationalizations (Six Classes)

Students employed 23 distinct moral justifications (R1–R23) for AI use, grouped into six Rationalization Classes:

• C1: Victimless Behavior (e.g., R1 No Human Victim, R2 AI-Synthesized Sources) — "copying AI-generated text is victimless" because no human is harmed; AI "doesn't have a soul."
• C2: Minimal AI Contribution (e.g., R3 Busywork, R5 Like Other Allowable Support) — the most common: AI is OK for "busywork" assignments deemed low-stakes, or is likened to human editors, grammar checkers, or writing centers.
• Ex Ante Contribution — students emphasize their own ideas, directions, or curation as decisive authorship evidence.
• Post Hoc Contribution (e.g., R10 My Paraphrasing, R11 My Verification, R12 My Style) — "any AI text reflecting their own beliefs or style is their own writing."
• Responsibility Denial / Normative (e.g., R15 Instructor Indifference, R16 Normlessness, R17 No Consequences, R18 Agency Denial) — externalizing responsibility and downplaying consequences.
• C6: Perceived Benefit (e.g., R19 Time Economy, R20 Educational Value, R21 Better Writing, R23 Better Outcome) — "I learn more by using AI extensively."

These rationalizations were ad hoc and post hoc, not self-consistent — students used them to explain behavior that had already occurred rather than to establish moral reasoning that informed behavior.

Belief vs. Behavior Mismatch

A striking finding is that continued AI use was not always driven by indifference or lack of ethical awareness. Some students were highly aware of the ethical stakes and experienced genuine moral distress (e.g., P16, who strongly wanted to obey a strict "never use AI" policy yet kept using AI, reporting that the prohibition "is creating conflict for me, because I'm breaking the rules"). A strict prohibition intensified internal moral conflict rather than preventing use — implying that punitive policy alone may backfire.

Authorship and Responsibility Made Contingent

AI-assisted writing lets students reinterpret harm, authorship, responsibility, and contribution in unusually flexible ways. Unlike traditional plagiarism (which involves a human victim), students can claim no victim exists. They claim ownership when AI output supports credit but distance themselves when it creates plagiarism or policy risks — externalizing responsibility in ways that echo moral disengagement (Bandura). This makes authorship contingent on context rather than stable.

What this means for practice

• Learners. Write out your own AI-use rule before a deadline forces the question, and check it against the course's formal policy — the study finds at least five sites where expectations diverge, and students' own normative views frequently depart from what the syllabus says.
• Learners. Notice which justification you reach for first ("no human victim," "it's just busywork," "the ideas are still mine") and test it deliberately: the 23 rationalizations catalogued here were ad hoc and post hoc, used to explain behavior rather than to guide it.
• Learners. Raise authorship questions with the instructor early. Students who wanted to obey a strict "never use AI" rule still used AI and reported genuine moral distress, so a prohibition on paper is not the same as a decision you can actually keep.
• Instructors. Teach ethical reasoning about authorship and harm explicitly, not just the rules: students' justifications were underdeveloped and self-serving, and writing clearer policy or escalating detection will not close a gap that opens at interpretation and practice.
• Instructors. Design assignments that make students account for their reasoning about contribution and harm, since purely prohibitive policies intensified moral conflict rather than preventing submissions.

Limitations

• The sample is 20 undergraduate students from 12 U.S. universities (15 female, 5 male), all self-reporting AI use at "sometimes" or higher on a screening survey, so students who avoid AI and broader populations are not represented.
• Instructors were never interviewed, so the authors cannot determine how faculty intended students to interpret or apply classroom AI policies — the "faculty intention" site is inferred rather than measured.
• The behavioral evidence is self-report plus uneven documentation: not all participants supplied their syllabi, submitted assignments, and AI logs, which the authors say may limit the ability to fully capture their practices.
• The findings describe a U.S. undergraduate writing context, and the authors interpret the rationalization taxonomy as a starting point requiring study in other writing-centered populations.

Citation

Kim, J., Toyama, K., Kim, S., & Carroll, J. M. (2026). "It's OK Because.": The Wild West of Student Rationalization of AI Use in Academic Writing. arXiv preprint.