Being accused of academic dishonesty is frightening under any circumstances. For a graduate student working on a thesis, it can feel like the accusation carries much more than a bad grade. Her entire degree could suddenly depend on whether someone believes she cheated.
That is exactly where one master’s student found herself.
Her professor and thesis supervisor had started using Pangram to check student writing for possible AI use. When one of her thesis chapters received a high AI-probability score, the professor treated the result as evidence worth taking seriously. The student insisted she had written the chapter herself and had months of drafts, notes, and version history to show how the work developed.
But her professor wasn’t convinced.
Then she found an older paper he had written on a subject close to her thesis. She managed to locate an unofficial PDF and decided to run it through the same detector.
The result was higher than the score on her own supposedly AI-written chapter.
So she sent it to him.
And suddenly, the conversation wasn’t just about AI detection anymore.
The Evidence Wasn’t Enough for Her Professor
The student explained that she had been working on her master’s thesis for months. Her professor, who was also her supervisor, had recently begun checking student writing with Pangram.
When he ran one of her thesis chapters through the software, the result came back with what he considered a high probability of AI involvement.
The student knew exactly what she was being accused of, and she had evidence ready.
She had earlier drafts. She had research notes. She had a complete version history showing the chapter evolving over time.
To her, this was about as close to a paper trail as she could reasonably provide.
Her professor disagreed.
According to the student, he told her that version history didn’t prove AI hadn’t been involved at some point. Pangram’s result was still evidence he had to consider.
That distinction terrified her because the professor wasn’t merely grading an assignment. He was her thesis supervisor, and she says he warned that if the issue couldn’t be resolved, he might not sign off on her thesis.
At that point, the student started investigating the detector itself.
That led to a rather uncomfortable experiment.
Then She Tested the Detector on His Paper
While researching how dependable AI detectors actually are, the student remembered that her professor had published a paper several years earlier on a topic closely related to her thesis.
She couldn’t find the paper through her university library anymore. Eventually, she located a PDF online through an unofficial source.
She admits she knew downloading it probably wasn’t entirely above board.
But instead of using the paper in her thesis, she used it as a test.
She ran the professor’s old paper through Pangram.
The result came back with an even higher AI probability than her own chapter.
That was the moment she decided to send the result to her professor.
Her message was essentially: This is the same software you’re using to evaluate my writing, and it’s saying your own paper is likely AI-generated.
The professor was not amused.
But what surprised the student was what he asked first.
He didn’t immediately argue about whether Pangram could produce false positives. Instead, he wanted to know where she had obtained the paper.
When she said she had found it online, he pressed for the exact website. Eventually, he told her that if she had obtained the paper from an unauthorized source, that was a separate issue that needed to be addressed.
Now the student was stuck with two different problems.
She still wanted her professor to address the reliability of the detector. But she was also worried that revealing exactly where she had downloaded the paper could get her into trouble.
She felt as though the conversation had suddenly moved away from the question she was trying to raise.
The Detector Question Is Actually Bigger Than One Professor
There is an important wrinkle here: AI detectors aren’t designed to function as magical lie detectors.
Several universities now explicitly warn faculty about relying too heavily on them.
Turnitin itself says its AI writing model can misidentify human-written material and that its AI report should not be used as the sole basis for adverse action against a student.
The University of Kentucky similarly explains that AI detectors produce probabilities rather than factual determinations and remain vulnerable to false positives and false negatives.
Some universities have gone even further. Caltech’s guidance strongly discourages the use of generative-AI detectors for academic dishonesty investigations, citing reliability, false positives, privacy, and fairness concerns.
And a 2026 report from Boise State University researchers concluded that current detectors have substantial limitations and that their outputs have significant constraints as evidence in academic-integrity decisions.
So the student’s central concern isn’t completely outlandish.
A detector score can raise a question. It doesn’t automatically answer it.
In fact, the University of California, Santa Barbara recently recommended that detection scores be treated as corroborating evidence rather than enough evidence by themselves to sustain an academic-integrity charge.
That makes the student’s drafts and version history particularly relevant. They don’t necessarily prove every detail of how she produced every sentence, but they provide contextual evidence about her writing process that a probability score alone cannot provide.
At the same time, the student’s decision to obtain her professor’s paper from an unauthorized source creates a legitimate separate concern.
Those two questions don’t cancel each other out.
The professor can reasonably ask how a potentially restricted paper was obtained. The student can reasonably ask whether an AI detector should carry enough weight to threaten her thesis.
The frustrating part is that both issues can be true at once.
Final Thoughts
The student’s biggest mistake may not have been running the professor’s paper through the detector. It may have been allowing the dispute to remain a one-on-one argument for so long.
The unauthorized download is worth taking seriously, and she shouldn’t pretend otherwise. But that doesn’t automatically settle the separate question of whether a probabilistic AI detector is reliable enough to threaten a student’s academic future.
If she genuinely wrote the thesis, her strongest defense isn’t a clever “gotcha.” It’s the paper trail: drafts, research notes, version history, correspondence, and a formal review of the evidence.
Sometimes the smartest way to win an argument with someone who has institutional power isn’t to argue louder.
It’s to make sure the argument is heard by someone else.
