Here is a sentence that should make every administrator who bought an AI-detection subscription wince: an AI tool accused a human of using AI, gave its verdict a confidence score of 100%, and was completely wrong. That is the heart of what happened to Orion Newby at Adelphi University, and in January 2026 a court agreed it was indefensible.

The accusation came from a machine

In the fall 2024 semester, Newby, a freshman in Adelphi's Bridges support program who is on the autism spectrum, submitted an essay for a "World Civilization" course. He had used tutoring support to write it. His professor, Micah Oelze, ran it through Turnitin, the plagiarism service that has bolted an AI-writing detector onto its older text-matching product. The detector returned an "AI-generated score of 100%." On the strength of that number, plus the professor's sense that the essay did not "carry the voice" he associated with a college student, Oelze gave the paper a zero and filed an academic integrity violation accusing Newby of using generative AI.

Sit with the mechanics for a second, because this is the part the special framing deserves. Turnitin's AI-writing detector is not a human reading a paper. It is itself a machine-learning system, an AI classifier trained to guess whether text was produced by a language model. The accusation against Newby did not originate with a person who caught him; it originated with an AI making a probabilistic guess and dressing it up as a hard "100%." The failure here is an AI tool producing a false positive. It is AI detecting AI, and getting it catastrophically wrong, then having a university treat the machine's hallucinated certainty as proof.

The evidence the school waved away

Newby denied using AI to write the essay and did the sensible thing: he brought counter-evidence. He submitted results from two other detection tools, ZeroGPT and Grammarly, that assessed the same essay as human-written. So the score sheet was one AI detector screaming "100% machine" against two AI detectors saying "this is a person." That alone should have detonated any notion that the tools produce reliable, dispositive truth. If three detectors look at one human-written essay and return wildly contradictory verdicts, the only honest conclusion is that none of them can actually prove who wrote it.

Adelphi was unmoved. On December 3, 2024, the university's academic integrity officer, Michael LaCombe, found Newby responsible and sanctioned him to a mandatory three-hour plagiarism workshop. The label "nondisciplinary" made it sound gentle, but a second offense could escalate to suspension or expulsion, so the finding put a loaded mark on his record. When Newby appealed, arguing he had not been given a fair chance to be heard, the same officer who issued the original finding denied the appeal.

The court was blunt

Newby took it to New York Supreme Court in Nassau County as an Article 78 petition, the mechanism for challenging arbitrary administrative decisions. On January 28, 2026, Judge Randy Sue Marber ruled in his favor in unsparing terms. The university's finding and its denial of the appeal were, in the judge's words, "without valid basis and devoid of reason," and the determination was arbitrary and capricious. The court ordered Adelphi to expunge the violation from Newby's record and rescind the sanction.

The judge zeroed in on the exact failure that should worry any school relying on these tools: the integrity officer "failed to even consider" the two AI detection programs that found the essay was human-written. The university, for its part, defended Turnitin in its filings as "reliable, accurate and an important tool." A court looked at that claim, looked at the 100% score that turned out to be false, and was not impressed.

Why these tools fail like this

AI-writing detectors share a structural problem: there is no ground truth in a block of prose. The detector cannot observe how the text was made; it can only pattern-match against what model-generated writing tends to look like statistically. That produces false positives at rates that are not remotely acceptable for accusing someone of misconduct, and the false positives fall hardest on predictable groups. Non-native English speakers, people who write in a plain or formulaic register, and neurodivergent students often produce exactly the kind of clean, structured prose these tools associate with machines. Newby, an autistic student who got tutoring help, sat squarely in that blast zone.

The "100%" is the most dangerous part. A score that precise reads like certainty to a busy professor or administrator, when it is nothing more than a confident output from a model that cannot actually know the answer. It is the same false authority that gets lawyers sanctioned for hallucinated citations, just pointed at students instead of judges.

Why this is a deployment failure, not just a due-process slip

It would be easy to file this under "school botched its disciplinary process," and the process was indeed botched: same officer ruling on the appeal, contrary evidence ignored, a vulnerable student steamrolled. But the chain of harm starts with an AI product malfunctioning. Turnitin's detector generated a false positive, the institution treated that machine output as dispositive, and a student paid for it until a judge intervened. The detector being an AI system that got it wrong is the load-bearing fact. Strip out the false "100% AI" verdict and there is no accusation to mishandle.

That is what lands it in the graveyard. An organization deployed an AI tool in a high-stakes role, the tool produced a confidently wrong result, and a human institution trusted it over the actual evidence. The grimly funny part is the symmetry: a student got hauled through a misconduct process because one artificial intelligence decided, with total confidence and zero accuracy, that his writing must have come from another artificial intelligence. The court's expungement order is the only thing in the story that got the answer right.