UVA School of Medicine (SOM) MD Program Policy on Generative AI

UVA SOM supports medical students to use generative AI and large language models to augment learning with appropriate educational use defined by faculty.

This policy is meant to accompany and help students better understand the University of Virginia’s Guidance, Responsibilities, and Terms of Use for the use of generative AI, which is available at https://in.virginia.edu/genai-useguidelines. Where the two policies overlap, the University-wide guidance takes precedence.

This document was created to define acceptable student use of generative AI software, including but not limited to ChatGPT, CoPilot, Gemini, Claude, DeepSeek, Perplexity, Elicit AI and Dall-E, and tools designed for medical practice, education, or research such as OpenEvidence, UpToDate Expert AI, and Consensus, etc. For purposes of this policy, we categorize these tools into two tiers:

1. Tools that are provisioned by UVA or UVA Health. With institutional data agreements these tools offer “safe harbor” for content. As of July 2026 – CoPilot, Google AI (including Gemini Notebook), Dax Co Pilot, and a small-scale pilot of Claude.
2. Content-Restricted External Tools. These typically require individual accounts and do not have institutional data agreements behind them. Some, such as OpenEvidence, Consensus, and UpToDate Expert AI, are designed for clinical queries and draw on the medical literature.

We acknowledge that there are many potential benefits of AI software as well as inherent risks (e.g., bias, limited contextual understanding, plagiarism, ethics and academic integrity, inappropriate language, errors in referencing, etc).

• Faculty will set expectations for the use of AI in their specific assignments including how the use of generative AI should be disclosed and cited.
• A distinction applies between AI-assisted and AI-generated work. AI-assisted work—where a student uses generative AI to support their own thinking (e.g., brainstorming, summarizing source material, or checking their own drafts)—generally carries a lower disclosure threshold than AI-generated work, in which the tool substantially produces the content submitted. Faculty should indicate, for a given assignment, whether and at what level of involvement disclosure is required; where faculty have not specified a threshold, students should err toward disclosure and describe how the tool was used. How do I cite generative AI in MLA style? (Updated and Revised) | MLA Style Center
• When submitting scholarly work for publication or presentation, students must adhere to generative AI policies set forth by journals and organizations and disclose when and how these tools have been used.
• Students are responsible for any inaccurate, biased, offensive, or otherwise unethical content they submit regardless of whether they personally authored it or used AI software to generate the content.
• Students may not create clinical documentation of any kind (e.g., H&Ps, progress notes, discharge summaries) within the authentic clinical environment using artificial intelligence applications outside of those supported by the EHR (e.g., Dax or Abridge).
• Students may not enter patient protected health information into any Content-Restricted External Tool, for any purpose. This includes not only clinical documentation but also clinical reasoning, question-asking, and image interpretation. Simply removing a name is not sufficient; the question must be framed generically and carry no identifiers such as name, medical record number, dates, images, or location. Reliable de-identification is harder than it looks—metadata, rare diagnoses, and identifiable features can still identify a patient. University Data Protection Standards | UVA Information Security
• The use of generative AI tools must be consistent with copyright, intellectual property rights, and relevant legislation (FERPA, HIPAA, etc.). See examples of acceptable use below.
• Licensed third-party content (e.g., UWorld questions or Anki decks built from copyrighted textbooks) may not be entered into any generative AI tool.
• If a student is found to have inappropriately utilized generative AI, the student will be subject to the UVA SOM Code of Conduct and greater University Honor Code.

Reviewing Recorded Course Content
A student uses Microsoft Copilot to summarize and highlight key concepts from the transcript of a recorded lecture in Panopto. Because this content is under UVA copyright, the student must use a UVA-provisioned tool with contractual data protection—no data leaves the institution, and no prompts are used for AI model training.

Creating Study Aids with UVA-Owned Content
A student uploads UVA course slides and faculty-generated notes from an immunology session to Gemini Notebook to generate Anki flashcards. Because these materials are UVA intellectual property, they may only be entered into a UVA-provisioned tool.

Generating Differential Diagnoses from a Classroom Case
During a small-group session in the Foundations of Clinical Medicine (FCM) course, students are presented with a fictional patient case. During and after class, a student uses ChatGPT to explore potential differential diagnoses and management plans based on the case. ChatGPT is a Content-Restricted External Tool. Because the case is simulated and contains no real patient information or proprietary UVA content, none of those content restrictions apply, so using ChatGPT is acceptable.

Using Generative AI to Answer a Clinical Question
A clerkship student is asked to present the answer to a clinical question during rounds the next day. The student enters the question into OpenEvidence or UpToDate Expert AI to find primary sources and understand the landscape of available evidence, reviews the most relevant primary sources themselves, and asks the AI clarifying questions to better understand the results. The student condenses this discussion into a brief summary to present during rounds. As the restriction on protected health information above requires, the student phrases the question generically with no identifiers, “What is the most likely differential for a geriatric man presenting with acute unilateral eye pain and blurry vision?”

Using Public or Student-Created Content with External Tools
To prepare for an upcoming exam, a student uses ChatGPT to generate quiz questions from publicly available online resources and personally written summaries of course topics (e.g., mechanisms of the adaptive immune response). Because these inputs are not UVA-owned or sensitive, they fall outside the restrictions that apply to Content-Restricted External Tools such as ChatGPT, so this use is permitted.

Using Generative AI to Understand Learning Objectives
A student copies a course learning objective into Mapify to generate a visual concept map and simplified explanation. Because learning objectives are general and not proprietary or sensitive, the use of this content-restricted external tool is allowed.

This Student Policy on Generative AI is subject to change to ensure alignment with UVA SOM, UVAHealth, and University of Virginia policies.

The Medical Education Management Committee has oversight and approval for this policy.

• Developed by the UVA SOM Generative AI Advisory Team and reviewed and approved by the Medical Education Management Committee. Updated July 2026.