Penn began piloting its artificial intelligence portal last week, featuring access to multiple large language models with an emphasis on data security.
PennChat, which includes certain Claude and ChatGPT models, started rolling out to the University community on July 7. The pilot is expected to conclude in mid-August with the launch of an official AI service.
According to Information Systems and Computing, PennChat is “hosted within Penn’s secure network environment” on Amazon Web Services “where companies cannot use Penn data to train AI models.” The service, which is based on the open-source tool LibreChat, provides access to Anthropic models through Amazon Bedrock and to OpenAI models through Microsoft Foundry.
Tiffany Hanulec, ISC’s executive director of technology services and chief technology officer, emphasized the importance of “secure access” to AI in a statement to The Daily Pennsylvanian.
“We believe generative AI will revolutionize the teaching, research, and business of Penn,” Hanulec wrote. “To reach that potential, we need to provide safe and equitable tool access to campus.”
For data protection, the interface’s LLMs cannot send queries to the internet and are only trained reliably on data collected before cutoff dates ranging from 2024 to 2025. ISC suggested that users ask the model they are using to learn its specific cutoff.
Although PennChat is “cleared for use with Low, Moderate, and most High Risk data” in accordance with Penn’s data classifications, ISC clarified that users should not enter Social Security numbers or credit card numbers and recommended against inputting protected health information.
Users must be connected to PennNet, AirPennNet, or the University’s GlobalProtect VPN to access the PennChat portal. Once on the platform, users can select from 14 different models, sorted into three categories based on credit consumption — premium, balanced, and economical — as well as a legacy category for older LLMs.
To prompt the models, each user is allotted a certain balance of credits that is refreshed daily if completely depleted. For at least some users, the balance is currently 1.85 million credits.
At a virtual town hall on Wednesday, ISC staff demonstrated PennChat’s capabilities and answered community questions about the platform.
PennChat users can create multiple chats and organize them in project folders. The interface also allows users to build custom agents by selecting an LLM, providing instructions, and toggling output parameters.
The platform supports file uploads and can generate Word, Excel, CSV, and PDF documents.
At the town hall, Hanulec explained that only a small group of ISC administrators with “trusted access” to the University’s sensitive information can view users’ chat history. She added that they have “no intention of looking at that data” outside of user-requested troubleshooting.
During the pilot phase, users can submit feedback through an online form. Hanulec wrote to the DP that PennChat will have information on “cost, capability, and data sensitivity level regarding different models” upon its full launch.
“PennChat is a new product and we are planning to continue building out the platform even after the August launch by adding new models and features as they become available for safe and approved use for the Penn community,” Hanulec wrote.
According to ISC staff at the town hall, PennChat could possibly feature the ability to share agents with other users in the future. Another area of consideration is connecting to external data sources such as PubMed.
The portal’s launch is part of a wider effort by ISC to provide AI tools, resources, and training. Penn community members already have access to Microsoft Copilot Chat through their University accounts.
Other universities have also moved to expand AI tools for their faculty, staff, and students. Dartmouth College and the University of Chicago recently launched institutional platforms for LLMs from Anthropic and OpenAI, respectively.
Staff reporter James Wan covers academic affairs and can be reached at wan@thedp.com. At Penn, he studies communication and computer science. Follow him on X @JamesWan__.
