Picture the defense: the committee has read the dissertation, and someone asks, almost casually, how much of chapter four you actually wrote yourself. A growing number of graduate programs now try to answer that question long before it reaches the defense room, requiring some formal accounting of AI use in a thesis or dissertation. The requirements are getting specific: named tools, named sections, sometimes a signature, rather than a general honor-code appeal to write your own work. What's notable is not just that these policies exist, but that most of them converge on the same structural pattern: a student and an advisor have to agree, in writing, on something that happened over months of writing that neither of them can independently check.
Pre-registration: ASU's applied projects and dissertations
Arizona State University's English department requires graduate students, starting Fall 2026, to submit a signed AI Agreement Form before starting the work:
"Starting in Fall 2026, graduate students must submit the AI Agreement Form before beginning work on an applied project, thesis, portfolio (written comprehensive exam), oral comprehensive exam, dissertation prospectus or dissertation. The form serves as a starting point for discussion and a record of agreement between the faculty member and graduate student."
ASU Department of English, AI policy
This is a pre-registration model: the terms of acceptable AI use are set before the writing starts, rather than described after the fact. It is not a zero-tolerance trap, either: "Non-compliance with the AI Agreement Form could create problems that considerably delay a student's progress, but it alone cannot be grounds for failure or dismissal" (ASU Department of English, AI policy).
Authorization-based: Rowan and UAB
Rowan University's interim academic integrity policy for graduate examination, thesis, and dissertation components frames permissible AI use narrowly:
"Permissible uses of AI may include grammar correction, formatting assistance, coding support, literature organization, ideation, or other limited supportive tasks when explicitly allowed by the instructor, advisor, or committee. Students remain fully responsible for verifying accuracy, originality, and appropriateness of all outputs."
Rowan University, interim academic integrity policy
Rowan also requires a disclosure statement wherever AI is used substantively, naming "the tools used, the nature and extent of use, and the sections or tasks affected" (Rowan University, interim policy).
The University of Alabama at Birmingham takes a similar approach, formalized in a standalone guidelines document rather than folded into the general academic integrity code: "AI use must be disclosed by the graduate student and approved by the capstone/thesis/dissertation committee and program director" (UAB, Guidelines on the Use of AI in Graduate Research).
Three institutions, three documents, one underlying model
| Institution | Model | What is required |
|---|---|---|
| Arizona State University (English) | Pre-registration | Signed AI Agreement Form before starting the thesis, dissertation, or prospectus |
| Rowan University | Committee authorization | AI use limited to specific tasks, only when the advisor or committee explicitly allows it |
| University of Alabama at Birmingham | Committee authorization | Formal written graduate-research guidelines govern what is permitted |
Across all three, the common thread is that failure to disclose, or use outside the authorized scope, is treated as a form of misrepresentation independent of whether the finished thesis is actually any good. A factually sound chapter written with undisclosed AI assistance beyond what was authorized is still a policy violation under all three frameworks.
The position this puts advisors in
A thesis or dissertation is typically the single largest document a student will have written up to that point, drafted over months, sometimes years. Whether the policy is pre-registration or committee authorization, it eventually asks an advisor to sign off on, or evaluate compliance with, something they were not present for and have no way to verify beyond the student's own account. That is a genuinely uncomfortable position for both sides, and it tends to surface for the first time at exactly the wrong moment: during the defense, when a committee member asks a pointed question about a section that reads a little differently from the rest.
This is the specific case where the general disclosure-statement problem becomes personal rather than abstract. A disclosure statement made by the person under evaluation, checked by the people evaluating them, with no independent record to consult, fails careful students exactly as often as it fails careless ones.
A shared record instead of a defense-room conversation
This is where writing the thesis in TeXposit changes the dynamic between student and advisor. Not by settling the authorization question (it still won't grade the thesis), but by giving both people something to look at together instead of only each other's word. The provenance report shows which chapters were typed directly, where an AI suggestion was accepted into the text, and a correlation score for how much of the student's own prose statistically echoes the project's AI chat log without having been generated by it, a more forgiving measure than a raw AI-word-count would be. For a pre-registration policy like ASU's, the same report retroactively shows whether actual use stayed inside the terms of the signed agreement, so nobody has to rely on memory to find out.
It has real limits. It doesn't establish intellectual contribution: a supervisor's idea typed out word for word by a student shows up as entirely student-authored text, because typing isn't thinking. It only covers text written inside the editor, so anything pasted in from elsewhere shows as a paste, not a traced source. And it has nothing to say about whether the argument itself holds up. What changes is not the judgment call, but whether there's a record to apply it to.