AI tools and data privacy: A University of Utah primer

All right, folks. It’s time to talk about artificial intelligence (AI).

Pop quiz: I’m a student at the University of Utah. I’ve heard the university has several approved AI tools, and I know some of my classmates use OpenAI ChatGPT Enterprise Edu to check their grammar before turning in essays or Google Gemini Chat to draft emails to professors. I have an exam coming up and want to use AI to help me study. What tools can I use without violating university policy, and how can I tell the difference between each tool?

Let’s break down the answer.

The university classifies data based on sensitivity. The first classification is restricted data, which includes personally identifiable information (PII) and protected health information (PHI). This data requires the most protection and can only be accessed or used for a job-related reason. The second classification is sensitive data, including employee and student information. Like restricted data, sensitive data may only be accessed by authorized individuals with an approved, job-related reason. The third and final classification is public data — essentially, any data that is publicly available, such as the university’s directory or bus route maps.

In our pop quiz, I’m specifically looking for help to study for my exam. I’m not sharing any PII, PHI, or student data. That means I’m working with public data. But what AI tools can I use with public data?

The top three university-approved AI tools are the U’s versions of Gemini Chat, Microsoft CoPilot Chat, and ChatGPT Enterprise Edu, all of which use large language models (LLMs) to generate responses to user prompts.

The U’s version of Gemini Chat is the preferred LLM for broad university use. It’s also approved for PHI use, as it meets U of U Health’s rigorous standards. Gemini is a closed-circuit system, meaning that user-input data is not sent outside of the university’s protected environment to train LLMs. The same goes for the U’s version of ChatGPT Enterprise Edu, which is best for campus, as it’s used to brainstorm ideas, revise written materials, and write basic code.

For general queries that draw only on public knowledge, there are several options, free and paid. For example, the U’s version of CoPilot Chat is free for students and staff, but those wishing to use Microsoft Teams Premium, which includes AI tools to assist with notetaking and collaboration, must purchase a license through the Office of Software Licensing. Other available tools include Scite.

Remember, personal subscriptions do not meet compliance requirements; to use these tools, you must log in using your university account. Unless you’re a healthcare worker using the U’s instance of Gemini Chat (specifically meant for use with PHI and PII), do not include any PII or PHI in your queries.

Final answer: I can use any of the university-approved AI tools with public data for appropriate academic reasons, such as studying, creating flash cards, or developing practice questions that draw on public knowledge. I should always double check the output to make sure it’s accurate.

Understanding how to use AI with different data types can be tricky. Luckily, there are several places to go if you need help or have a question.

The Information Privacy Office handles data privacy and protection for the U of U and U of U Health. We’re always happy to answer privacy and AI questions, and if we don’t know the answer, we’ll do our best to route you to folks who can help. Additionally, the AI Leadership Team reviews new AI tools and use cases on a regular basis. Keep up with the latest guidance and tools for campus use at ai.utah.edu and U of U Health use at the AI Health SharePoint site (login required).

 

Node 4 Search