All posts
11 min read

Universities That Disabled AI Detection — Verified Tracker (2026)

Vanderbilt, Waterloo, Yale and Curtin have switched off Turnitin's AI detector, each confirmed on the institution's own page. Widely repeated "50+ universities" counts are not verifiable — here is the difference, with sources.

Four universities have disabled Turnitin's AI writing detection with documentation on their own websites: Vanderbilt (August 2023), the University of Waterloo (September 2025), Yale (current), and Curtin University (1 January 2026). You will see claims elsewhere that more than 50 institutions have done the same. We tried to verify that figure and could not, so this tracker separates what is confirmed from what is merely repeated.

That separation is the point of this page. Anyone appealing a flag, setting policy, or reporting on this trend needs to know which claims survive contact with a primary source.

We should disclose our position: we sell a tool that rewrites AI-generated text, so we benefit commercially when readers doubt AI detectors. That is exactly why this page holds itself to institutional primary sources and marks our own uncertainty openly. Where a widely circulated claim favors our commercial interest but cannot be verified, we say so rather than repeat it.

Key takeaways

  • Vanderbilt University disabled Turnitin's AI detector in August 2023, citing opacity and calculating that a 1% false-positive rate implies roughly 750 wrongly flagged papers out of ~75,000 annual submissions. [1]
  • University of Waterloo discontinued the feature as of September 2025, announced by the Associate Vice-President, Academic. [2]
  • Curtin University disabled Gen-AI detection from 1 January 2026 while keeping text-matching active — the two are separate decisions. [3]
  • Yale University's Poorvu Center states that "the AI detection feature of Turnitin is currently disabled," without giving a date or reasoning. [4]
  • The frequently cited "more than 50 universities" figure could not be verified against institutional sources, and neither could most of the individual names attached to it. [5]
  • Disabling AI detection is not the same as permitting undisclosed AI use — Curtin retained its integrity procedures in full. [3]

What this page covers

  1. The verified tracker
  2. Why these institutions decided as they did
  3. The claims we could not verify
  4. What this means if you have been flagged
  5. What this means if you set policy
  6. Honest limitations, then a FAQ and full sources

The verified tracker

Every row below is sourced to the institution's own page or official announcement. Last checked 13 August 2026.

InstitutionCountryStatusEffectivePrimary source
Vanderbilt UniversityUSAI detector disabledAugust 2023Vanderbilt Brightspace announcement [1]
University of WaterlooCanadaAI detection discontinuedSeptember 2025Office of the AVP, Academic [2]
Curtin UniversityAustraliaGen-AI detection disabled; text-matching retained1 January 2026Curtin official notice [3]
Yale UniversityUSAI detection feature disabledNot statedPoorvu Center, Turnitin page [4]

Four confirmed institutions is a smaller number than this topic usually advertises. It is also a number you can check yourself in about ten minutes, which the larger ones are not.

Why these institutions decided as they did

Vanderbilt: the arithmetic of a 1% error rate

Vanderbilt's August 2023 statement remains the most detailed public reasoning any university has published, and its central move was to convert a percentage into a headcount. Turnitin claimed roughly a 1% false-positive rate. Vanderbilt submitted about 75,000 papers in 2022. That implies around 750 students wrongly flagged in a single year. [1]

Its other objections: Turnitin gave no detailed account of how determinations are made, describing only that the tool looks for patterns common in AI writing without specifying them. The feature arrived with under 24 hours' notice and, at the time, no way to switch it off. And in more than one internal test, the tool flagged entirely human-written text as 100% AI-generated. Vanderbilt's Instructional Technologies team judged the advantages inconclusive and raised documented bias against non-native English speakers as an equity concern. [1] That equity concern is the best-evidenced part of the case against these tools — we cover why AI detectors flag non-native English writers and the research behind it separately.

Waterloo: a decision taken across faculties

Waterloo's Associate Vice-President, Academic announced in May 2025 that AI detection would be unavailable to university users as of September 2025. The published notice emphasizes consultation — Undergraduate Associate Deans, AFIW Academic Deans, the Undergraduate Operations Committee, Work Integrated Learning, Cooperative and Experiential Education, and the EdTech Steering Committee. [2]

Secondary coverage attributes to Waterloo an internal finding that the tool flagged human-written work as 100% AI-generated. That detail appears on a companion institutional page rather than the announcement we verified, so we cite it with that caveat rather than as confirmed from the primary notice. [2][5]

Curtin: the middle path

Curtin's approach is the one most often misreported, and the distinction matters. From 1 January 2026, across all campuses and study periods, only the Gen-AI detection feature was switched off. In Curtin's own words, "the regular text-matching originality checks will remain active." [3]

The university framed the change as fostering "trust and clarity within a modern academic culture" and ensuring assessments are "secure, fair, relevant and future-ready." Academic integrity procedures continue unchanged, and students are still directed to follow assessment instructions on appropriate AI use. [3]

If you are drafting policy, Curtin is the useful template: it separates detection of AI from detection of copied text, and from the integrity process itself. Those three are routinely collapsed into one another in public discussion.

Yale: confirmed, unexplained

Yale's Poorvu Center for Teaching and Learning states on its Turnitin page that "the AI detection feature of Turnitin is currently disabled." No date, no reasoning. [4]

Secondary sources attribute to Yale a rationale about false-positive rates being incompatible with the burden of proof in integrity proceedings. We could not find that on Yale's own page, so we are not presenting it as Yale's stated position. [5]

The claims we could not verify

This section exists because the alternative is repeating things that sound useful and happen to suit us.

"More than 50 universities." This figure circulates widely, usually spanning the US, UK, Canada, Australia and South Africa, and often appears on sites selling humanizers or detection-adjacent products. We found no underlying list with institutional citations. Unverified. [5]

"At least 12 elite institutions," naming Johns Hopkins, Northwestern, Georgetown, Pittsburgh and NYU. Reported by aggregators. We could not confirm these from institutional pages. Some may well be accurate — Northwestern is reported to hold that detection results alone are insufficient for a misconduct finding, and Johns Hopkins to treat detection as advisory only, both of which are plausible and consistent with the wider trend. Still unverified. [5]

University of Cape Town (October 2025) and University of Arizona. Reported, not confirmed by us. [5]

Specific quoted rationales attributed to Yale and Johns Hopkins. Appear in secondary coverage; not located on institutional pages. [5]

A note on why this matters practically. If you cite "50+ universities" in an appeal and the panel asks which ones, an answer you cannot substantiate weakens everything else you said. Four institutions you can link to is a stronger argument than fifty you cannot.

What this means if you have been flagged

Useful, with limits. This tracker demonstrates that serious institutions, having tested these tools on their own students, judged them unfit to carry an integrity case. That is context a panel should hear, and it sits on top of the published accuracy research we summarise in do AI detectors actually work? and the dated figures on our detector test results page.

It is not, by itself, a defence. Your institution's decision to keep using a detector is not overturned by another institution's decision to stop. What actually clears people is process evidence — version history, dated drafts, notes, and a substantive conversation about your own argument. Institutional context supports that; it does not replace it. Our separate guide on what to do when you are falsely accused of using AI covers the first 24 hours and ranks the evidence by how much weight it actually carries.

One thing to avoid: do not run disputed work through a humanizer, including ours. It breaks the correspondence between your submitted text and your version history, which is the evidence that clears you, and it looks like tampering.

What this means if you set policy

Three observations from the confirmed record, alongside our fuller guide to what an AI detection score can and cannot support:

Convert the percentage to a headcount. Vanderbilt's most persuasive argument was arithmetic. Apply your institution's annual submission volume to your vendor's published false-positive rate and look at the resulting number of students. [1]

Decide the three questions separately. Whether to detect AI, whether to detect copied text, and how integrity cases are adjudicated are distinct. Curtin switched off the first, kept the second, and left the third intact. [3]

Consultation is doing real work here. Waterloo's notice reads as a record of a process rather than a technical judgement. [2] Decisions arrived at across faculties tend to survive contested cases better than decisions imposed by one office.

Honest limitations

  • This tracker is deliberately incomplete. It lists what we could verify, not everything that has happened. Institutions change policy quietly, and some genuine cases are certainly missing.
  • Absence of evidence is not evidence of absence. That we could not verify Johns Hopkins or Northwestern does not mean those reports are wrong — only that we would not assert them.
  • Policies change and pages get edited. Every row was checked on 13 August 2026. Verify before relying on any of it in a formal proceeding.
  • Yale's entry has no date, so it cannot be placed in the timeline with the others.
  • We benefit if you distrust detectors, which is why this page is built on institutional primary sources and marks unverified claims as unverified. Read it with that in mind.
  • This is not legal or institutional advice. If you face suspension, expulsion, or visa consequences, speak to a student defence attorney.

FAQ

How many universities have disabled Turnitin's AI detection? Four are confirmed here from institutional sources: Vanderbilt, Waterloo, Yale and Curtin. Widely circulated counts of 50+ could not be verified. [1][2][3][4][5]

Did Curtin stop using Turnitin altogether? No. Only the Gen-AI detection feature was disabled from 1 January 2026; text-matching originality checking remains active and integrity procedures are unchanged. [3]

Why did Vanderbilt disable it? Lack of transparency about how determinations are made, the roughly 750 papers a year implied by a 1% false-positive rate at its volume, internal tests returning 100% AI on human-written text, and documented bias against non-native English speakers. [1]

Does my university still use AI detection? Check your institution's teaching-and-learning centre or academic integrity pages, which is where all four confirmed decisions were published. Policies vary by institution and sometimes by course.

Can I cite this page in an appeal? You can cite the primary sources it links to, which is stronger. Link the institutions' own pages rather than this article.

Does a university disabling detection mean AI use is now allowed there? No. These are separate decisions, and Curtin's notice makes that explicit. [3]

Is Turnitin removing AI detection entirely? No. Individual institutions are switching the feature off; that is a customer configuration decision, not a product withdrawal.

Sources

  1. "Guidance on AI Detection and Why We're Disabling Turnitin's AI Detector," Vanderbilt University Brightspace, 16 August 2023 — the fullest public reasoning any university has published, and the source for the 750-papers calculation, the transparency objection, and the internal 100%-AI test results. Cited because it is the institution's own account, not coverage of it.
  2. "Waterloo discontinuing the use of AI detection tool Turnitin.com," Office of the Associate Vice-President, Academic, University of Waterloo — official announcement, cited for the September 2025 effective date and the consultation record. The primary notice, rather than press summaries of it.
  3. "Update on Turnitin AI-Detection Tool," Curtin University — official notice, cited for the 1 January 2026 date and the verbatim confirmation that text-matching remains active. The distinction between the two features comes directly from this page.
  4. Poorvu Center for Teaching and Learning, Yale University — Turnitin instructional-tools page — cited only for the verbatim statement that the feature is disabled. No reasoning is attributed, because the page supplies none.
  5. Aggregator and secondary coverage (AI Weekly, GradPilot, AIPlagGuides, SupWriter and similar) — the origin of the "50+ universities" figure, the "12 elite institutions" list, and the quoted rationales for Yale and Johns Hopkins. Listed explicitly as the unverified tier so readers can see exactly which claims rest on it. Several of these sites sell products in this category, ourselves included, which is a reason for caution in both directions.

Try it on a paragraph — 500 words a day free, no account needed.

Open the editor