How do you cope with the problem of not exposing sensitive project data when using tools like ChatGPT or Claude to automate basic project tasks (minutes of meeting, slides, ecc) ?

Michele -

Unfortunately, there's only two choices - your company could subscribe or purchase a secure, non-public instance of a given Gen AI tool so that real data can be used OR you scrub data and use public instances.

Kiron

In my experience, the type of ai your speaking of deals with transcribing meeting notes with ai technology such as Otter.AI. There are a few ways that a company could secure sensitive data with this. First, the notes are there as evidence of meeting goals. There may be some team members that need a reference to those goals. Typically there should be only 1-2 people tasked with transcribing the meeting notes, and these members should be the ones with access to secured transcription platforms. I know lately transcribing with ai has expanded to apps like Microsoft 365 and certain Chat GPT apps and now even Nvivo. For those apps that I am familiar with, team versions allow individualized permission settings. Therefore, the app itself can be set so that only the designated transcription personnel have access to the original notes. If transcribing with an app like 365, copilot sets it where the transcriber can draw out the financial data to place into a permanent format. Then ask the app to write over all numerical data with a symbol such as ###### or $$$$$$ to signify a monetary amount. Then, the transcriber can use a list of names to have the app cover over names with a symbol or a word such as name. Some trasncription apps also allow a setting where the "script" created says person 1, person 2, etc rather than to name the person as it's transcribing. If the app doesn't easily pull out numberical or fianncial data, the tanscribed version can be pulled through an app like Nvivo or R studio (possibly python but it's mostly numerical bound) and even Tableau to ask the app to remove the data and draw over it with a symbol since these apps will just look at data characteristics without trying to understand it unless asked to understand it, and then follow through with a command. This takes an extra step, but higher versions of 365 or even Adobe do well to look for characters, remove and replace. Also, the final notes can also be password protected and set with layered permissions with the correct centralized IT department structure that houses personnel familiar with administrating layered permissions.

Does this help at all?

Michele -

Unfortunately, there's only two choices - your company could subscribe or purchase a secure, non-public instance of a given Gen AI tool so that real data can be used OR you scrub data and use public instances.

Kiron