Harvard designs AI sandbox interaction without compromising security

[twno1]

https://news.harvard.edu/gazette/story/newsplus/harvard-designs-ai-sandbox-that-enables-exploration-interaction-without-compromising-security/

[twno1]

In my understanding, in order to obtain results, data still needs to be sent to services like OpenAI and Azure... etc.

But they said "The prompts and data entered by the user are viewable only to that individual; they are not shared with LLM vendors, nor can the data be used to train those models."

How is that achieved?