Document, Document, Document: Authorship Provenance as an Eligibility Criterion in Public Education

Abstract

The literature on artificial intelligence detection is a literature about detectors, asking whether classifiers are accurate and whose writing they misread. Underneath that contest sits a remedy on which nearly everyone agrees. Keep your version history. Compose inside a tool that logs your process. Be able to show your work. Integrity offices, vendors, journalists, and the writers who attack detection most sharply all give the same advice, and all of them offer it as neutral prudence. This Article is about the advice. Process-logging instruments authenticate authorship by inspecting how text entered a document. Grammarly Authorship sorts every character into a fixed taxonomy: typed by a human, generated by artificial intelligence, modified or edited with it, or pasted from a known or unknown source. There is no category for text that was spoken. Google Workspace for Education, the dominant platform in American K-12 schooling, ships no authorship tool at all; its Originality Reports check for copied text rather than machine authorship, and the record teachers actually consult is Google Docs version history, which logs a large insertion with a timestamp and supplies no label for it. Dictated text arrives as a block. It is scored as an unattributed paste, or left unlabeled for a human to interpret privately. The research program underwriting these instruments states the defect in its own vocabulary, because educational measurement researchers have built keystroke classifiers separating original composition from reproduction, and the published feature sets enumerate insertions, deletions, pauses, and pastes with dictation nowhere in the list. An accommodation is a different process. That is what an accommodation is, and in a school the difference is legally required, individually determined, and written down by the entity that will later evaluate the work. This Article makes three claims. First, that process-based provenance measures input mechanism and reports the result as authorship, and that published demonstrations of trivial evasion establish the gap rather than suggesting it. Second, that a documentation requirement applied as a condition of academic evaluation is an eligibility criterion under 28 C.F.R. 35.130(b)(8) and must be shown necessary, which is already settled law where documentation burdens fall on disabled students. Third, that the school case is the strongest configuration of this problem anywhere in the detection literature, because the defendant is a public entity, the individualized education program closes the knowledge element, Perez v. Sturgis Public Schools clears exhaustion, and Payan preserves the one category of damages surviving Cummings.

Keywords

• civil rights
• education law
• Section 504 Rehabilitation Act
• ADA Title II
• screen out
• deliberate indifference
• keystroke logging
• writing assessment
• academic integrity
• speech-to-text
• assistive technology
• individualized education program