Answers › False positives

Can autistic or neurodivergent students be wrongly flagged by AI detectors?

Updated 30 August 2026 · 4 min read
Short answer

Yes — and it's documented, not theoretical. In 2026 the Office of the Independent Adjudicator upheld a complaint from an autistic student accused on the strength of a detector score and later cleared.

Detectors score predictability, so consistent structure, precise repeated terminology and even sentence rhythm all raise the number. That's a property of the tool, not evidence about the writer.

Why these writing patterns score high

A detector asks one question of every word: how expected was that, given what came before? Writing that is internally consistent produces a lot of expected words. Several features common in autistic and ADHD writing — and several explicitly taught by university study-skills services — do exactly that:

The double bind

Structured writing templates are frequently recommended through disability support and study-skills provision. A student can follow the strategy their university gave them and be flagged by software the same university runs.

Where the evidence stands in 2026

HEPI published twice over the summer of 2026 on the fairness gap in AI detection, concluding that a detector score should never be sufficient evidence of misconduct on its own. The OIA cases that year involved both non-native English speakers and a neurodivergent student. The pattern across them is the same: the tool produced a number, the number was treated as evidence, and the finding didn't survive scrutiny.

The wider reliability picture is the same one everyone faces — a Stanford study found roughly 61% of essays by non-native English speakers wrongly flagged — but the groups affected overlap and compound.

If you're accused

The general steps apply — ask what the allegation is, request the report, freeze your drafts, involve your students' union. Two additions specific to this situation:

Keeping drafts and version history in one continuously-edited document remains the single most useful protection, because it replaces an argument about writing style with a record of the work happening.

Know what your writing looks like to a detector

SafeGrade shows you which passages read as machine-like and why — so nothing arrives as a surprise. Private, never stored, never used for training.

Run one free scan →

No card. No catch. First scan free.

Sources: Higher Education Policy Institute — AI detectors and the fairness gap (August 2026); HEPI — Catching the wrong students (July 2026); Stanford study on AI-detector false positives for non-native English speakers; Office of the Independent Adjudicator case decisions, 2026. SafeGrade reports third-party research and does not publish its own detection-accuracy figures.