I would draw the line at authorship, accountability, and transparency—not at a percentage generated by an AI detector. AI-assisted research should remain acceptable when AI is used as a tool to support activities such as brainstorming, language editing, literature organization, coding assistance, or exploring ideas, provided that the researcher independently verifies the output and remains intellectually responsible for the final work. Misconduct begins when AI effectively replaces the researcher’s own intellectual contribution—for example, fabricating data, inventing references, generating analyses that the researcher cannot understand or defend, or submitting AI-produced work while deliberately misrepresenting it where disclosure is required.
The difficult part is that AI detectors cannot reliably establish where that line has been crossed. Current guidance from universities such as Oxford, UBC, and Penn State cautions against treating detector scores as evidence of misconduct because false positives and false negatives remain significant. The example of a pre-AI thesis being flagged therefore raises a fundamental question: if the instrument cannot reliably identify AI use, should it determine whether someone has committed misconduct?
In my view, the line should be established by academic institutions and disciplinary communities, through clear, transparent policies developed with researchers, educators, students, and research-integrity experts. Individual instructors should apply those rules consistently, while academic-integrity committees should make formal misconduct determinations based on evidence and due process—not an algorithmic score. Ultimately, the strongest test is whether the researcher can explain, defend, verify, and take responsibility for the work. AI may assist the process, but intellectual accountability must remain human.
