← All education signals
Trust signal

The Detector Doubt

An AI detector gives a probability that a machine wrote the text. Treating that number as proof risks losing in court, as a New York court showed in January 2026.

What is it?

An AI-detector score is losing standing as proof of cheating. A New York state court annulled a finding built on a Turnitin AI score because counter-evidence was ignored, and Curtin University has switched its detector off. The score becomes one input to test, not a verdict.

Why now?

The Adelphi ruling came on 28 January 2026 and was reported within weeks. Curtin’s switch-off took effect on 1 January 2026, after an announcement the previous September. Both land as universities decide how to police AI writing without trusting a single number.

Who’s building it?

Integrity panels, university assessment teams, detector vendors, student advocates and the courts reviewing sanctions.

Curtin UniversityTurnitinIntegrity panelsState courtsStudent advocates
The full read
Why now, who is building it, where the market goes, and what stays unsettled.
Get weekly →
01 · The market
Who’s in it
Builders — detector vendors and the universities that configure their tools.
Buyers — students flagged by a score, and the panels deciding their cases.
Gatekeepers — courts reviewing sanctions and the university procedures they test.
What’s at stake
Whether a classifier score can carry a sanction on its own, or only open an inquiry weighing other evidence.
Whether universities keep paying for AI detection at all.
Whether records of drafting replace the score as proof.
What’s unsettled
Whether courts outside New York treat a detector score the same way.
Whether other universities follow Curtin, or keep the tools on.
What evidence of authorship a panel should accept when detectors disagree with each other.
02 · How the demand matureda progression, not a timeline
Novelty
Detectors promise a quick number for a question markers cannot answer by eye.
Retention
Panels lean on the score, and students answer with rival detectors and drafts.
Willingness to pay
A family spends six figures to clear one record, and a court sides with them.
Category
Proof of authorship moves from a score to a record of how work was made.
03 · Where the market goesthe forks, not a prediction
→The tested score: detectors stay, but every flag triggers a hearing where the student’s drafts and rival results are weighed.
→More universities switch AI detection off and keep only text-matching and authorship records.
→Courts elsewhere back institutions, and the Adelphi ruling stays a lesson in procedure, not evidence.
→Detector vendors publish error rates, and panels set thresholds they can defend.
The forks turn on whether institutions can show how a piece of writing was made, cheaply enough that nobody needs to guess from its finished text.
04 · The Assessment deskadjacent market, moving now
The Supervised Page
Written work moves back under watch.
Ofqual asked exam boards for immediate steps against AI misuse.
Schools Week
The Human Mark
Human rescoring lifts many essays that machines marked.
About a third of Dallas ISD’s resubmitted tests scored higher.
The Austin Chronicle
The Bias Audit
Algorithms that judge people now face outside tests.
Colorado repealed its bias-audit AI Act in May 2026.
Colorado General Assembly
05 · The Wire
01
A NY court annulled a Turnitin-based finding.
NY Supreme Court (FindLaw) ↗
02
Two other detectors had said the essay was human-written.
Inside Higher Ed ↗
03
Curtin University disabled Turnitin’s AI writing detection from 1 January 2026.
Curtin University ↗
04
Family sues Palo Alto school district over a 76% Turnitin AI score
SF Standard ↗
The panel on this signal
They agreed on
–The court faulted how the university used the score, not the detector’s accuracy, so the ruling concerns process.
–One university switch-off is a record, not a trend count.
–A lone detector score is weaker evidence than a drafting record.
The forensic linguist disagreed on
–Authorship is judged against a writer’s own known samples; a detector scoring a text against a population answers a different question.
Know more about the method →
✦How this signal was built
01 · Panel
A Delphi panel of 4 AI analysts, each reasoning independently.
→
02 · Debate
They challenge and revise each other’s reads over two rounds.
→
03 · Evidence
Every claim is tied to a cited, linked source on The Wire.
→
04 · Synthesis
A moderator writes it into one read, disagreement and all.