Vanderbilt's Brightspace administrators posted a public statement the week before the fall semester began. The post explained that Turnitin had rolled the AI detector out without a beta period, without documentation of the underlying model, and without a clear appeals process for flagged students. Vanderbilt's team had asked Turnitin for methodology detail and received the vendor's marketing collateral in reply.
Quick answer
Vanderbilt disabled Turnitin's AI detector in August 2023 citing false positives, methodology opacity, and equity concerns. It has not re-enabled the feature. Several peer institutions have since followed a similar path. Turnitin still ships the detector by default to institutional customers. The core reasoning behind the Vanderbilt decision has held up.
To understand why the retrospective matters, it helps to revisit the exact language Vanderbilt used, then trace what the sector has done since.
The 2023 decision
The disabling was not framed as a rejection of AI detection in principle. It was framed as a rejection of a specific tool that had shipped without the disclosure a research university expects from an evaluative instrument. That framing matters, because it did not commit Vanderbilt to a permanent no-detector stance. It committed the institution to a standard of evidence.
Vanderbilt did not reject AI detection. It rejected a tool that had shipped without the disclosure a research university expects.
The Working Educators editors
What Vanderbilt said then
The 2023 statement made four specific claims. First, that a 1% false positive rate at the document level, applied to a campus that submits tens of thousands of papers, produces hundreds of wrongful flags per term. Second, that the flags were being generated without a way for students to inspect the evidence. Third, that the burden of disproving a machine's judgment fell on the student. Fourth, that AI-generated text and human writing under time pressure look statistically similar to the classifier.
None of these claims required a lab study to make. They were arithmetic and procedural. The Washington Post's coverage that same month walked readers through the pattern of students being asked to prove a negative. Vanderbilt's post was the institutional response to that reporting.
What has changed at Vanderbilt since
The detector has stayed off. Vanderbilt's Center for Teaching has instead published guidance on assignment redesign, oral follow-ups, and AI use disclosure. Faculty who want detection scores can still run essays through third-party tools on their own account, but the institution does not do it centrally and does not treat those scores as evidence in academic integrity cases.
The shift has moved integrity work upstream. Faculty spend more time designing assignments that reward course-specific reading and revision, and less time adjudicating detector scores after the fact. That is a real cost in course prep, but it also produces a different kind of writing to grade.
Which institutions followed
Michigan State's teaching center reviewed the tool and issued cautious guidance that stopped short of endorsement. Northwestern faculty raised similar concerns in departmental meetings covered by student media. Several community college systems in California and Oregon paused institution-wide adoption pending better disclosure. None of these were as public as Vanderbilt's post, but the pattern is consistent: institutions with active teaching-and-learning centers have generally not adopted detector scores as adjudicative evidence.
The exception is administrative pressure. Some registrar and general counsel offices have pushed for detector deployment as a documentation shield in disputes with parents and accreditors. Where that pressure has succeeded, the deployment has usually come with caveats limiting how scores can be cited in integrity hearings.
What Turnitin kept shipping
Turnitin still ships the AI writing indicator by default. The current FAQ (updated October 2026) reports a document-level false positive rate below 1%, and a Turnitin blog, June 2023 put the sentence-level rate near 4%. Both come from Turnitin's internal testing. The methodology behind those numbers has not been published in a form outside researchers can replicate.
The company has published a false positive explainer and additional support material. Neither addresses the Vanderbilt objection directly, which was not that the numbers were wrong but that the disclosure was not sufficient for institutional adoption.
The reasoning that aged well
The Stanford group's study on TOEFL essay false positive rates and Liang et al.'s finding of a 61.3% false positive rate on non-native English writing came out in the same window as Vanderbilt's decision. Krishna et al.'s paraphrase evasion work followed. Each has since been cited in the wave of institutional reviews that have quietly retired classroom detector policies.
The methodological alternative that has emerged looks different from Turnitin's approach. Proofademic's published research paper discloses its academic-population false positive rate at the sentence level and describes the classifier's training approach. That is not proof of superior accuracy in every setting, but it is the kind of disclosure Vanderbilt asked for in 2023 and did not get.
Frequently Asked Questions
Has Vanderbilt re-enabled Turnitin's AI detector?▼
No. Three years after disabling it in August 2023, Vanderbilt has not re-enabled central access. Faculty can still use third-party tools on personal accounts, but the institution does not deploy or treat scores as adjudicative evidence.
Did other universities follow Vanderbilt's example?▼
Several did, though usually with less public statements. Michigan State issued cautious guidance, Northwestern faculty raised concerns internally, and several community college systems in California and Oregon paused adoption. The pattern is consistent among institutions with active teaching-and-learning centers.
What was Vanderbilt's main objection?▼
The core objection was methodological opacity, not the abstract idea of AI detection. Turnitin did not publish enough about how the classifier worked for Vanderbilt to defend integrity cases built on the tool's scores.
What did Turnitin change in response?▼
Turnitin published additional support material, including a false positive explainer, and continues to report internal accuracy figures. The company has not published external replication studies of the classifier.
What replaced the detector at Vanderbilt?▼
Assignment redesign, oral follow-ups, and disclosure norms. The Center for Teaching has published guidance on rubrics that reward course-specific engagement, which is harder to produce with a general-purpose model.
The Bottom Line
The Vanderbilt decision looks better in 2026 than it did in 2023. The underlying research on false positive rates, the paraphrase-evasion literature, and the pattern of quiet institutional withdrawals all point in the same direction. The tool has not improved enough, or disclosed enough, to reverse the decision.
The teaching-side alternative has also matured. Assignment redesign, oral defense, and disclosure norms are more work than running a detector, but they produce a different kind of student writing and a stronger record when an integrity case does need to be built.
The lesson for other institutions is not that AI detection is unusable. It is that a detector adopted without external methodology disclosure will not stand up to the kind of scrutiny a good faculty senate applies. Vanderbilt asked the right question first.
