If you are a college student, it is likely that you have heard of AI detectors being used to detect generative artificial intelligence and large language models (LLMs) in papers. One of the most popular AI detection tools is Turnitin, used by over 16,000 institutions worldwide, including Brookdale Community College.

Students worldwide are at risk of being falsely flagged for generative AI and may face consequences such as failing, being suspended, or even expelled due to false flags in their papers. AI-detection tools used by institutions—such as those built into Turnitin—are too unreliable to justify the academic penalties that they can trigger. Yet, students are often treated as guilty based on opaque and unproven technology.

Most readers (including faculty and administrators) do not understand how AI detection tools work. In fact, AI detectors like Turnitin do not know whether the text was written by AI or not. They estimate probability based on patterns like sentence length, vocabulary and predictability in papers. These patterns are not exclusive to AI. Human writing, especially academic writing, can often look like generative AI. When these detectors rely on writing style rather than verifiable evidence, they are guessing, not proving.

The unreliability of Turnitin and other AI detectors matters because of the harm that false positives can cause. Students have already reported being flagged for AI despite writing their work independently (even at Brookdale). In many cases, the detector’s output percentage or confidence score—is treated as authoritative, even though companies like Turnitin themselves acknowledge that the results are not always accurate, and results should not be the sole basis for adverse actions against a student.

Once they are flagged, students are often placed in a difficult position of having to prove their innocence, reversing the expectation that accusations must be supported by evidence. A probability score—though appearing sophisticated—is not proof.

The unreliability shown by AI detectors has already alarmed numerous institutions, some banning the Turnitin tool altogether. Huge universities, including UC Berkeley, Colorado State, Georgetown, Indiana, MIT, NYU, Northwestern, Syracuse, Yale, University of Maryland, and University of Texas at Austin, have all banned Turnitin’s AI detection software. New York University (NYU) put out a statement about their reasoning for disabling the AI tool on Turnitin, stating that it doesn’t work well enough to confidently apply it, and later going on to state that while Turnitin claims a 1 percent misidentification rate, their work showed that rate to be closer to 4 percent, meaning that 1 in 25 student papers would be marked as AI-written.

The dangers that students face when AI detectors are used in academic integrity violations are massive. Academic penalties can carry lasting consequences, from failing grades to disciplinary records that affect scholarships, transfer applications, and graduate school opportunities. In high-stakes situations, due process matters. Students are rarely, if ever, told how a score was generated, what specific features of papers were flagged, or how likely their tool is to be wrong. When AI detectors function as arbiters, students lose the ability to challenge accusations made against them.

Beyond just individual cases, widespread reliance on AI detection risks reshaping the entire educational environment. When students know their writing will be evaluated by AI detection software designed to search for suspicion, students do not focus on developing ideas—they worry about how their writing will be interpreted by a mere algorithm. Faculty, however, are already coming to rely on automated judgments rather than engaging directly with a student’s work. Education is shifting from mentorship and trust from students to surveillance and enforcement, which should not be the case.

Earlier, I mentioned the rate NYU found of a 4 percent misidentification rate, but AI-detection tools also disproportionately flag certain groups of students—students that Brookdale has a lot of: multilingual writers, neurodivergent students, and those who follow the academia that is taught to them in introductory courses. If a system penalizes deviation from an assumed “normal” writing pattern, it risks reinforcing existing inequalities in writing. A technology that cannot distinguish between style differences and misconduct should not be used to judge academic integrity.

To be clear, concerns about misuse of AI are valid. Students should not be sending AI-generated work as their own, and faculty deserve tools to help uphold academic standards. However, convenience cannot replace evidence. A tool with known limitations should not be treated as an authority, especially when conclusions are probability-based, and its inner workings are inaccessible.

Flawed technology does not protect academic integrity. Trust, transparency and fair standards of evidence do. When probability is mistaken for proof, students suffer. The credibility of education suffers as well.

I have spoken with multiple faculty members about this issue, most agree that this is a recognized problem. However, because they don’t have another system to be put in place, the known flawed system is continuing to be used. This is a huge injustice to students who pay tuition and fees to be in these classes.

If you or someone you know wants to speak up to the administration regarding this issue, please contact me at [email protected]. It is time for a change to be made.

DISCLAIMER: This article was written with zero assistance from generative AI platforms and any large language models. At the conclusion of this article, it was run through GPTZero AI Detection (with advanced scanning), where it received a 27 percent AI-generated score.