AI and Law School Academic Integrity

Two documents control, in this order: your law school's honor code (or code of academic integrity) and the individual course syllabus. The honor code supplies the offense categories — typically plagiarism, unauthorized assistance, fabrication, and false statements to the school. The syllabus supplies the permission. Most codes were drafted long before ChatGPT and say something like "a student shall not give or receive unauthorized assistance," with the professor defining what is authorized. That drafting choice is why the answer varies course to course inside the same building.

This genuinely varies by school and by professor, and the variation is driven by three things. First, whether your school issued a default rule after 2023 — some adopted a school-wide presumption that AI is prohibited absent express permission, others left it entirely to faculty, and a few adopted a permitted-with-disclosure default. Second, the assessment type: in-person closed-universe exams, take-home exams, seminar papers, journal write-on competitions, and moot court briefs are often governed by separate, stricter rules. Third, the professor's pedagogical view, which is why you can be permitted to use AI for a first-year legal writing brainstorming exercise and expelled for the same conduct on a seminar paper.

Do not reason from what your classmates say another professor allows. Read the syllabus language on AI, and if it is silent, email the professor and get a written answer before you use anything. That email is also your best evidence later.

Some conduct violates essentially every honor code regardless of how permissive your school is, because it involves misrepresenting authorship, using prohibited aid on an assessment, or putting false material in front of a reader.

The hard cases involve tools that are AI but do not feel like it. Grammarly's grammar checker and Word's editor have generative features layered on top; the distinction most policies draw is between correcting your sentences and generating new ones. Westlaw's and Lexis's AI-assisted research and summarization features sit in the same space — several schools expressly exempt research-database tools from their AI bans on the theory that they are research, not writing, but many say nothing at all. If your policy is silent on embedded database tools, ask.

The other gray zone is process use that never appears in your submission: asking a chatbot to explain the Erie doctrine, generating practice hypotheticals, having it quiz you on your outline, or asking it to critique the structure of a draft you wrote. Most faculty treat pure study use as equivalent to a commercial outline or a tutor, and most policies are aimed at submitted work product. But a professor who has banned AI "in connection with this course" has written a broader rule than one who has banned "AI-generated text in submitted work." Read the verb.

Translation and accessibility uses are a third category worth flagging to your professor or disability services office in advance rather than assuming.

Not primarily through AI detectors. Detector output has a well-documented false-positive problem, particularly for non-native English writers and for heavily edited formulaic legal prose, and most honor code bodies will not sustain a charge on a detector score alone. Some schools have instructed faculty not to use them as evidence at all.

What actually generates charges is process evidence. Exam software logs application switching and time stamps. Word and Google Docs retain version history showing whether a document was drafted or pasted in a block. Fabricated or nonexistent citations are self-proving. Sharp discrepancies between a student's cold-call performance and their written product prompt a professor to ask questions, and the answers a student gives in that conversation frequently become the case. So does a first draft that quotes a case that does not exist and a final draft that quietly drops it.

The practical consequence: keep your drafts, your research trail, and your notes. A student who can produce dated outlines, downloaded cases, and successive drafts almost never loses an AI accusation. A student with a single final file has nothing to show.

This is the part students underestimate. Bar applications in every jurisdiction ask whether you have been the subject of academic discipline, investigation, or sanction in law school, and the question is typically not limited to findings that appear on your transcript. Your law school also completes a certification of character to the admissions authority. A single honor code violation is rarely disqualifying by itself; jurisdictions admit applicants with far worse in their history after full disclosure and evidence of rehabilitation.

What is disqualifying is nondisclosure. Model Rule 8.1 prohibits knowingly making a false statement of material fact in a bar admission application and prohibits failing to disclose facts necessary to correct a misapprehension. Applicants get denied for concealing minor incidents far more often than for the incidents themselves. If you are charged, assume you will be disclosing it, retain every document, and get the school's written record of the outcome.

Also expect a delay. Character and fitness review with a disclosed academic discipline item routinely adds months, which can mean starting a job unlicensed.

The academic rules track the professional ones closely enough that treating law school as practice is the right instinct. ABA Formal Opinion 512 (2024) addresses lawyers' use of generative AI and grounds it in existing duties: competence under Model Rule 1.1 and Comment 8's technology competence obligation, confidentiality under Rule 1.6 (which means not feeding client information into a tool that trains on inputs), supervision under Rules 5.1 and 5.3, candor to the tribunal under Rule 3.3, and reasonable fees under Rule 1.5 when AI reduces the time a task takes.

The verification duty has teeth. In Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023), the court sanctioned lawyers who filed a brief containing citations generated by ChatGPT to cases that did not exist. In Park v. Kim, 91 F.4th 610 (2d Cir. 2024), the Second Circuit referred an attorney to its grievance panel after she cited a nonexistent decision produced by AI. Courts have continued issuing sanctions on the same facts. There is no defense that the tool made it up; signing the filing is the representation.

Practically, the habit to build now is that you verify every authority in the reporter or on Westlaw or Lexis, read the case, and confirm it says what you claim. Do that on every 1L memo and the professional rule takes care of itself.

Adopt a documented protocol and use it consistently. Ask permission in writing and save the reply. Keep a research log and dated drafts. Never paste an unreleased exam, a professor's problem set, a journal write-on packet, or anything under a confidentiality condition into a public tool — that can be a separate violation even where AI use is permitted. Read and Shepardize or KeyCite every case before it goes in a document with your name on it.

On citation form, the Bluebook has no dedicated rule for generative AI output, and practice is unsettled. Where disclosure is required, the safe approach is to describe the use in a footnote or cover note — what tool, what version, what date, and for what task — rather than to invent a citation format. Law journals increasingly have their own AI disclosure policies for authors and editors; ask your journal's articles editor rather than guessing.

Finally, calibrate to the stakes. AI is a reasonable study aid for generating hypotheticals and testing your understanding. It is a poor substitute for the analytical work that exams actually measure, and using it to produce your outline instead of writing one forfeits the learning that outlining exists to create.