AI Plagiarism Detection: Why We Don’t Ship It
If you came here comparing AI-detection features, here is the unusual answer: OpenEduCat does not offer AI-writing or plagiarism detection — deliberately. The evidence on false positives and bias convinced us that detector verdicts cannot carry the weight schools put on them. This page explains the evidence, and what we build for academic integrity instead.
What the detector evidence shows
Three findings, each independently disqualifying for disciplinary use. Verify them — they are all public record.
Detectors produce false accusations at scale
AI-writing detectors output probabilities, not proof — and at institutional volume, even a small false-positive rate accuses real students falsely every week. OpenAI withdrew its own AI-text classifier citing low accuracy, and universities have publicly walked back detector-based enforcement after wrongful-accusation cases. A tool that occasionally convicts the innocent is not a tool a school should automate discipline on.
The bias lands on your most vulnerable students
Published research (including a widely cited Stanford study) found AI detectors flag writing by non-native English speakers at dramatically higher rates — formulaic-but-honest prose reads as "AI-like." The students most likely to be falsely flagged are exactly the ones least equipped to contest it.
It is an arms race the detector loses
Paraphrasing tools and newer models defeat detectors faster than detectors improve. Building integrity policy on a technical control that decays is building on sand; building it on assessment design and process is durable.
The integrity approach we build instead
Assessment design beats detection
Process-visible assessment — drafts, oral defenses, in-class components, personalized prompts tied to course discussion — makes undisclosed AI use both harder and less rewarding. The platform’s assignment workflows support staged submissions and version history, so instructors can see work developing rather than judging a single artifact.
Clear AI-use policy, enforced as process
The workable integrity stance is a clear policy on when AI use is permitted and disclosed — see our AI policy checklist — combined with assignment-level expectations. Our platform gives staff policy chat grounded in your own handbook, so the rules students are held to are the rules everyone can look up.
Sanctioned AI reduces unsanctioned AI
Students with a governed practice-and-study AI (adaptive quizzes, cited course chat) inside the course portal have less reason to take assignments to consumer chatbots. Visibility and guardrails on the sanctioned path shrink the shadow path.
Human adjudication, with real evidence
When a case does arise, it should rest on evidence a person can defend — version history, process artifacts, a conversation with the student — not on a probability score. Our audit-log architecture applies to AI use on the platform, so what the sanctioned tools did is always reviewable.
Frequently Asked Questions
Common questions from academic integrity officers and IT leaders.
Integrity by design, not by classifier
See staged submissions, policy chat grounded in your handbook, and governed student AI in a demo — the integrity stack that doesn’t accuse anyone falsely.
We would rather lose a checkbox comparison than sell a probability score as proof.