Professor Wei Liu had been at the university for eleven years. She had published twenty-three peer-reviewed articles, supervised nine doctoral students, and served on four departmental committees. Her tenure review - the culmination of over a decade of academic work - was scheduled for spring 2026. The dossier she submitted in January included her five most significant publications, her teaching evaluations, her service record, and letters from external reviewers confirming the quality and originality of her scholarship.
One member of the tenure committee - a recently hired associate professor in a neighboring department - decided to run Liu's publications through an AI detection tool. It was not required by university policy. It was not discussed with the committee chair. It was a unilateral decision, made possible by the easy availability of detection tools, and it would derail Liu's tenure review for three months.
The Flags
Three of Liu's five submitted publications were flagged. The scores ranged from 38% to 56%. The committee member brought the results to the next meeting and placed them on the table. "I think we need to discuss these," they said.
The publications in question had been written between 2019 and 2023. Two of them predated ChatGPT's release. All three had been through peer review, published in respected journals, and cited by other researchers in the field. The idea that they were AI-generated was, on its face, absurd. But the numbers were on the table, and numbers have a way of commanding attention regardless of context.
The Investigation
The committee chair, to her credit, recognized the situation's complexity. She paused the tenure review and consulted the dean's office, which consulted the provost's office, which consulted the university's legal counsel. The consensus was that the AI detection results, while not part of the formal tenure process, could not be simply ignored once raised. A supplementary review was initiated.
Liu was informed of the flags in a meeting she describes as "surreal." She was shown the detection reports and asked to provide evidence that her publications were her own work. For someone who had spent eleven years building a body of scholarship - years of research, writing, revision, peer review, and publication - the request was both procedurally reasonable and personally devastating.
"They asked me to prove I wrote papers that have my name on them, that went through double-blind peer review, that are cited in other people's work," Liu told us. "Papers I wrote before the technology they're accusing me of using even existed. How do you respond to that?"
The Evidence
Liu assembled her defense with the thoroughness that characterized her academic work. She provided: original manuscript files with metadata showing creation dates; email correspondence with co-authors and journal editors spanning years; reviewer comments and her responses; conference presentations where she discussed the work in progress; and her research notebooks, maintained in both English and Mandarin.
She also retained an independent expert who analyzed the flagged publications and explained the false positives. Liu's academic writing style - formal, structured, and precise, reflecting both her disciplinary norms and her training in English as a second language - produced the low-perplexity, low-burstiness patterns that detectors associate with AI output. The expert's report noted that academic writing in the hard sciences routinely triggers false positives at rates significantly above the population average.
The Resolution
The supplementary review concluded after three months. The committee found no evidence of AI use and cleared Liu's publications. The tenure review resumed and Liu was ultimately granted tenure. The university issued no formal statement about the incident.
Liu describes the experience as "a stain that doesn't wash out." The three months of investigation delayed her tenure by a semester, disrupted her research schedule, and created a period of uncertainty that affected her ability to supervise her doctoral students effectively. The committee member who initiated the detection scan faced no consequences.
The Larger Problem
Liu's case raises a question that universities have barely begun to address: what happens when anyone with access to a free detection tool can unilaterally trigger an investigation of someone else's work? The barrier to accusation is essentially zero - paste text into a website, get a percentage, raise a concern. The barrier to defense is enormous - months of evidence gathering, expert consultation, and emotional labor.
This asymmetry is the structural flaw at the heart of the current system. Until institutions establish clear policies about when and by whom detection tools may be used, and what evidentiary standards apply to the results, cases like Liu's will continue. The tools are available to anyone. The consequences fall on the accused.