Columbia University

Under Columbia University Standards and Discipline 2026–2027 (§C.3), adopted at the discretion of the Dean of each adhering school, absent a clear statement or approval in writing from a course instructor granting permission, the use of generative Artificial Intelligence (AI) tools to complete an assignment, exam, or any other coursework is prohibited, including using AI for translation services, crafting study materials, organizing a works cited page, or polishing a student's work.

Under Dean's Discipline procedures (§J), a student found responsible for a policy violation may appeal the outcome or sanction within five (5) business days after receipt of the decision/outcome letter; an initial screening is conducted by the AVP (or designee) within 5 business days; if returned for exceeding the 5-page maximum or failing to meet the three exclusive grounds, the student may resubmit within two (2) business days from return (one-time resubmission); once grounds are met, the appeal is sent to the Dean (or designee) who has 10 business days from the date of receipt to deliver a response; and the student will receive the Appellate Officer's final written decision within 10 business days of receipt of request or at their earliest convenience (extensions communicated in writing per footnote 7).

Under Columbia University's Provost Generative AI Policy, a report from an AI detection tool confirming suspected AI use cannot be used as the sole piece of proof in a Dean's Discipline case to find a student responsible for a violation of policy.

When an instructor suspects a student of using generative AI in violation of course policy, the instructor should refer the student to the Dean's Discipline process rather than attempting to resolve the matter independently.

Under Columbia Law School's default academic policy (§9.2, subject to instructor written alternatives under §Default and §9), all work submitted for academic credit must be student-composed and reflect the student's independent intellectual creativity and judgment unless expressly indicated otherwise by the course instructor in writing; revising or paraphrasing AI text does not transform it into student-composed work, and translating one's own non-English writing into English to submit as original work is prohibited by default (§11).

Columbia Business School policy establishes that when faculty permit generative AI in a course, students will be instructed how to properly use and cite such tools in their work; failure to properly and completely cite AI-generated responses may be reported as a violation of the CBS Honor Code.

Under Columbia Law School Policy§7, students must disclose any AI use contributing to a written submission or other work product; students should be transparent and err on the side of disclosure and follow additional instructor citation requirements.

Under Columbia Law School's 2026–2027 Policy (§Default), the rules set forth in Sections 8 through 11 are considered default academic policies; individual course instructors may adopt alternative course- or assignment-specific policies as long as they do so in writing, at a minimum in the initial course syllabus, and with advance notice to all students.

Under CUIMC policy, users of CU AI Toolkit are approved to use Sensitive Data (PHI) only when using approved models (currently OpenAI models within the toolkit); dual-affiliates must note that full chat history is visible to all models used in a chat, and therefore must not input Sensitive Data unless all models in the chat are approved for Sensitive Data; furthermore, Live Web Search through the OpenAI API is strictly unapproved for Sensitive Data.

Under CUIMC data classification rules, while FERPA student education records are generally classified as Confidential Data, elements containing student identifying or health information are PII or PHI classified as Sensitive Data; when this overlap occurs, the stricter Sensitive Data classification and tool restrictions apply rather than standard Confidential Data rules.

Columbia University Information Technology (CUIT) governs institutional AI deployment through a centralized Data Classification Table, requiring all community members to verify approved data classifications before using AI services, while directing CUIMC members to specialized health data policies.

Columbia researchers, faculty, and graduate students must avoid uploading unpublished research data, unpublished scholarly manuscripts, funding proposals, or research subject PII into public or unvetted AI tools, and are strictly prohibited from using generative AI for peer review; sensitive research analysis must be restricted to CUIT-approved secure environments under IRB protocols or CUIMC HIPAA-approved platforms under IRB/AIGC review.