Fulcrum Solutions Let’s talk

Blog

Is it safe to put customer data into ChatGPT or any AI tool?

It depends on what you put in and what you use it for. Anonymise first, use the right plan, keep names out where you can, and know the privacy rules. Andrei on the safe way and the unsafe way.

Andrei Busu1 min read

Illustration in film noir style: a robot in a trenchcoat and fedora leans against a brick wall in a rainy alley, holding out a manila dossier to a man in a suit who reaches to take it.

It depends on what you put in and what you use it for.

A lot of what you'd put in can be anonymised first. The AI usually doesn't need a customer's name, phone number or address to do the job, so you take those out before anything goes in. If you are dealing with sensitive or identifying information, there are ways to handle that too, but it takes more care.

Illustration: the same trenchcoated robot at a desk under a lamp, reading a fanned stack of handwritten pages with lines blacked out at random and an anonymous blank-faced photo clipped to the top page.
Take the name, number and address out before the file goes anywhere. The job itself is all the AI needs.

It also depends on which kind of tool you're using. A free chat window is one thing. Going through an API, or a business plan, is another: the providers say they won't train on your data by default, and some let you choose where it's stored, including in Australia. Those Australian options are more limited and restrict what you can build, so I'd rather not use them unless a client needs the data to stay here.

Infographic titled Three ways in, three columns: Free chat window with the OpenAI logo, may be used for training, you choose nothing about storage, fine for anonymised work; API or business plan, not used for training by default, some let you choose where data is stored including Australia, what we build on for clients; Your own server, data never leaves your business, model runs on hardware you control, more to set up and maintain. Footer: Australian-hosted options exist but limit what you can build.
A free chat window, an API or business plan, or your own server. The plan decides what the provider may do with what you send.

For most businesses and most people, it isn't a problem. But privacy law adds a few considerations, and they're worth knowing about before you start. They're too much to cover properly here, so I'd point you to the Office of the Australian Information Commissioner, which has plain- English guidance.

And you don't have to use ChatGPT at all. You can run your own model on your own server, so customer data never goes to a third party.

So the safe way is to decide what goes in, use the right plan, and keep names out where you can. The unsafe way is staff pasting whatever they like into a free chat.

Illustration: the robot waits under a street lamp across a rainy street from a building signed Information Bureau, its front door wide open and unguarded, shelves of dossiers visible through the door and windows.
A free chat with no plan behind it is an open door. Decide what goes in before someone is waiting outside.
ShareLinkedInXEmail

Keep reading