UT Spark: Usage & Data Privacy Guide

UT Spark: Usage & Data Privacy Guide


Welcome to UT Spark: UT’s internal platform for exploring generative AI. This guide explains how your data is handled when using the platform and provides key considerations to help you use it responsibly.


Your Data Stays Private

  • Your data is not used to train external AI models and is not shared with third-party vendors.

  • UT has implemented controls to ensure that content you submit remains private and secure.

  • While interactions are logged for security and support, they are not publicly viewable or reused.

  • All prompts and conversations within UT Spark (nebulaONE) are stored within UT's environment. 

For details, see the official Cloudforce privacy statement: https://gocloudforce.atlassian.net/wiki/spaces/NQ/pages/844529699


Web Search

What is Web Search for an AI Agent?

Web Search is a feature that enables an AI agent to access and retrieve information directly from the internet in real time. Instead of relying only on its built-in knowledge, the agent can perform live searches, review trusted sources, and incorporate the most up-to-date information into its responses.

Why is Web Search Important?

  1. Keeps Information Current – AI models are trained on data up to a certain point in time. Web Search allows the agent to overcome this limitation by pulling in the latest updates, news, and research.

  2. Expands Knowledge Beyond Training Data – Some questions may fall outside of what the model has seen during training. Web Search allows the agent to look up new or niche information it otherwise wouldn’t know.

  3. Improves Accuracy and Reliability – By citing and referencing sources, Web Search helps ensure responses are backed by verifiable information, making the AI’s output more trustworthy.

  4. Supports Specialized Use Cases – Whether it’s academic research, technical troubleshooting, or checking institutional policies, Web Search equips the agent to handle domain-specific or time-sensitive tasks effectively.

Caution: Agents with Web Search

Some Official Agents may have web search enabled. When this is turned on:

  • Your prompts (or parts of them) may be sent to external search engines to gather results.

  • This means data leaves UT’s secure environment during that specific interaction.

If you're working with confidential, sensitive, or unpublished information, avoid using agents with web search enabled.

Official Agents using Web Search feature:

  • (OpenAI) ChatGPT-4o

  • (OpenAI) ChatGPT-4.1

Disabling the Web Search feature in personal agents

When creating or editing your personal agent, you can disable the “Search the Internet” capability by toggling the feature off (as shown below).


What’s OK to Input?

While the UT Information Security Office’s Cloud Services Decision Matrix does not yet specifically call out UT Spark (nebulaONE), controlled data is generally considered safe for use within the platform's current implementation of OpenAI models. Users should continue to exercise caution and follow institutional data handling guidelines.

Feature

How Your Data Is Handled

Feature

How Your Data Is Handled

Web Searches

Private, not stored or shared by Brave Search or Bing Search

External AI Models

Encrypted, not used for training, privacy-respecting

Prompt Storage

Per-user, secure Azure database, access-controlled

Deleted Chats

Redacted, not fully deleted, files stored separately

Data Location

All within your organization’s Azure environment, only accessible to database administrators


Have Questions or Feedback?

Please fill out the feedback form!