A running list
AI detection lawsuits, tracked
Every case we can cite where a student challenged an AI detector finding in court, with the detector, the score, and what happened. Updated as records change.
In short
As of September 2026, one court has rejected an AI detector finding on its merits: Newby v. Adelphi University, where a Turnitin 100 percent flag was found without valid basis and expunged. Other cases, including Kato v. Palo Alto USD, Doe v. Michigan, Rignol v. Yale and Yang v. Minnesota, have turned on due process, and one expulsion was affirmed.
Why this list exists
Students accused on the strength of a detector score keep asking the same question: has anyone fought this and won. The answer is yes, a few times, in narrow ways, and the details matter more than the headline. This list holds only cases with a court record or a docket we have checked, plus one petition included for context. Where our records do not carry a detail, the cell says so rather than guessing.
The cases
| Date | Case | Detector and score | What happened | Source |
|---|---|---|---|---|
| January 28, 2026 (context) | University at Buffalo student petition | Not a single case; campus AI detection generally | Petition reached 1,437 signatures. The same NBC News report described students using humanizers to avoid false flags, with 33.9 million visits to 43 humanizer sites in one month. | NBC News, January 28, 2026 |
| Early 2026 | Newby v. Adelphi University, New York Supreme Court, Nassau County, 2026 NY Slip Op 26021 | Turnitin, 100 percent; Grammarly and ZeroGPT read the paper as human | The court found the university’s finding “without valid basis” and ordered the record expunged. The one decision so far that rejected a detector-based finding on its merits. | Justia |
| February 2026 | Doe v. University of Michigan | Not in our verified records | Disability-related claim. Further detail is not in our verified records. | GradPilot tracker |
| February 2026 | Yang v. University of Minnesota | Not in our verified records | Expulsion affirmed. The student did not prevail. | GradPilot tracker |
| May 5, 2026 (filed) | Kato v. Palo Alto Unified School District, U.S. District Court, N.D. Cal. | Turnitin, 76 percent | The family filed a 1,162-page packet including Google Docs revision history and seeks an order barring detector scores from being treated as dispositive evidence. The district answered in June 2026. Pending. | N.D. Cal. docket |
| Pending (filing date not in our records) | Rignol v. Yale University | GPTZero; score not in our records | Pending. | GradPilot tracker |
Source: Justia (Newby); N.D. Cal. docket (Kato); GradPilot tracker (Doe, Yang, Rignol); NBC News, January 28, 2026 (Buffalo petition); all checked 2026-09-08.
How to read this list
Read the outcomes column before the score column. Adelphi is the only case where a court examined the finding itself and rejected it, and even there the ruling was that the finding lacked a valid basis, not that the detector was wrong as a matter of fact. The other cases have been argued on process: whether the student got notice, saw the evidence, and had a hearing. Winners so far won on due process, not on disproving the detector. Yang is a reminder that when an institution’s process holds up, the student can lose. If you are facing an accusation, the practical lesson is to insist on every procedural step, in writing, and to bring your own record of the work: drafts, notes, version history, and where one exists, a keystroke-level writing record that shows typed versus pasted and the order in which the argument formed. The step-by-step page walks through it.
A record of the writing session is evidence for that conversation, not a verdict. It shows how text entered the editor and nothing about whose ideas they were. None of the cases above turned on such a record, because none of the students had one. That is worth knowing before the next assignment.
The numbers behind the cases
The context for all of this is scale. Turnitin reported on February 24, 2026 that 15 percent of submissions between October 2025 and February 2026 were more than 80 percent AI, against 3 percent at the detector’s launch in 2023. A Guardian freedom of information request found about 7,000 UK academic misconduct cases involving AI in 2023 to 2024. Against that volume, a false positive rate that Turnitin itself states as under 1 percent at the document level still produces a steady stream of students with a wrong number next to their name. The false positive rates page does the arithmetic.
Source: Turnitin press materials, February 24, 2026; Guardian FOI, 2024; turnitin.com false positives page; checked 2026-09-08.
Add to this list
If you know of a filed case, a decision, or a correction to a row above, send the docket or the citation through the contact page. We add rows only with a source we can check, and we date every change.
Questions people ask
Has any court ruled that AI detectors are unreliable?
Not in those words. In Newby v. Adelphi the court found a finding built on a Turnitin 100 percent flag to be without valid basis, which is a ruling about that finding, not about detectors as a class. Kato v. Palo Alto USD asks a federal court to bar detector scores as dispositive evidence, and that request is pending.
Do students usually win these cases?
The record is small and mixed. The wins so far have come from the institution skipping a step in its own process. Yang v. University of Minnesota went the other way, with the expulsion affirmed in February 2026. Most accusations never reach court and are resolved inside the institution.
Is this list legal advice?
No. It is a reading list with dates and sources. If you are facing an accusation, read your institution's policy, use its student advocacy office, and consult a lawyer only if an internal appeal fails and the consequence is severe.
If you write on the record
Ask your instructor about EssayGuard.
When a class writes in EssayGuard, every student has a replay of how their paper was written, and nobody has to argue with a percentage. Faculty can start for free.