A student accused of using AI says he ran his professor’s own published paper through the detector, with a telling result

AI detectors have become standard equipment in university misconduct disputes, and one master’s student decided to point the software back at the person using it. According to the Daily Dot, the student posted to Reddit on September 3 after their supervisor accused them of using AI in a thesis chapter. The post has drawn more than 3,100 upvotes.

The professor had been running student writing through Pangram, a commercial AI detector, and one of the student’s chapters came back with a high AI-probability score. The student says the chapter was written without AI and handed over months of notes, drafts, and version history as proof. The professor allegedly countered that version history could not rule out AI at some stage, and suggested the dispute could affect thesis approval.

That is when the student went looking for the professor’s own work. Unable to get a paper on the same topic through the library, they found a PDF from what they described as an unofficial source and ran it through Pangram. The result came back with a higher AI-probability score than the flagged chapter.

The professor’s response was about the PDF, not the software

Rather than engage with the detector’s reliability, the professor reportedly focused on where the student had obtained the paper. Commenters seized on that, arguing it dodged the actual question. Others advised the student to stop handling it informally and escalate to the department chair, the graduate program director, or the academic-integrity office.

The account has not been independently verified, and the people and institution involved remain unidentified. It’s important to note that the student’s test was also not a clean experiment. Pangram warns that PDFs can cause parsing errors and recommends raw text or .docx files, as formatting, references, and technical material can all skew results.

The timing of the professor’s paper is also unknown. Some commenters noted that a paper published a few years ago could still have been drafted or revised after AI tools became widely available. However, those caveats do not resolve the underlying problem, which is that detectors produce probabilities rather than proof of authorship.

Pangram’s own numbers make the point. The company reports that its latest model, Pangram 4, has a false-positive rate of 0.0041%, or roughly one false flag per 24,000 documents. Its technical report also states plainly that both false positives and false negatives do occur, and that the model’s predictions are to some degree a black box.

The company’s guidance for educators says a high score should not be treated as automatic proof of cheating. Pangram suggests talking to the student about their process, inspecting the flagged sections, and looking for corroborating evidence. That framing makes the student’s version history considerably more relevant than the score itself.

Turnitin has taken a comparable position, stating its AI detector should not be the sole basis for action against a student. The company also stopped displaying precise percentages below 20% because of false positives in that band. Human academic writing frequently shares the traits detectors look for, including formal transitions and standardized phrasing.

The dispute also sits inside a much wider shift in how students actually use these tools. Research has documented teens turning to ChatGPT without telling anyone, which is part of why institutions have leaned so hard on detection software.

The student did not prove the professor wrong or the detector broken, which wouldn’t be the first time. What the test may have done is make it harder to treat a single score as evidence, which could leave the documented writing process as the only defense that holds up.