AI Cheating Detectors Are Flagging the Students Who Need Support Most

One student’s accusation, one district’s reckoning

Eleanor Canina, a Green Hope High School freshman, received a zero on an English assignment after her teacher ran it through three AI detection programs, each returning scores between 62 and 87 percent for AI generation. Her family appealed. A second teacher used version history to trace every keystroke from the original draft and confirmed she had written every word herself. The grade was changed to a 100.

Canina then took the case to the Wake County school board, speaking publicly and launching a petition calling for responsible AI detection policies. The district responded by drafting a new policy that explicitly bans AI detection programs, stating the district does not support their use due to ‘technical unreliability, inaccuracy, and potential for bias against specific student populations, including those for whom English is a second language.’ The policy also requires students to disclose AI use and makes unauthorized use academic misconduct, with teachers, not algorithms, as the final reviewers. A board vote is expected in August or later.

The tools are broken, and the students they break are predictable

This is not a story about one imperfect tool in one district. A 2026 study evaluating commercial AI detectors on a balanced dataset found false positive rates ranging from 43 to 83 percent on authentic student writing. A separate 2026 analysis found a mean false positive rate of 61.3 percent for essays written by non-native English speakers, compared with 5.1 percent for native speakers. The detector does not see writing quality. It sees writing patterns, and it flags the ones that differ from the neurotypical norm.

That is a direct problem for students with dyslexia, ADHD, and autism. Researchers documenting false positive patterns found that neurodivergent students face elevated detection rates precisely because their natural writing, which often includes repetitive phrasing, consistent word choices, and non-standard syntax, overlaps with what detectors were trained to catch. Grammarly, a common accommodation for students with dyslexia, also increases detection flags. The tool punishes students for using the support the system already recommended they use.

The pattern is familiar. Researchers debunked learning styles in 2008; a 2020 review found nearly nine in ten educators still taught to them. Schools adopted AI detectors without evidence they work, and the students who pay the price are the ones who were already navigating extra barriers. That is not a technology problem. It is a systems problem.

What parents need to ask before it becomes a problem

Ask your child’s school whether they use AI detection tools. If they do, ask what the appeal process looks like, and whether tools like Grammarly or speech-to-text, which many students with learning differences rely on, are factored into any review. The Canina case resolved because version history existed and a second teacher reviewed it. Not every school has that infrastructure, and not every family knows to ask for it.

A detection score is not evidence of misconduct. It is a flag from a tool with documented error rates up to 83 percent on authentic work. If your child is accused, request the specific scores and the tool used, ask for human review by a teacher who knows your child’s writing, and request access to version history if available. The system Wake County is moving toward, where teachers are the final word and detectors are removed entirely, puts the right person in that role: the adult who actually knows your child.