This weekend, I was talking to someone who mentioned frequently using generative AI for their work but, being well aware of the risks, always making sure they never ever pasted any personal or client information into the chatbot. They’d read the stories and were a responsible guardian of the data.
A little later in the conversation, they mentioned using Claude to scan all the files on their laptop’s drive and sort them based on a series of criteria.
I asked if Claude could read the file contents. “Of course” they answered, “that’s how he knows what to do with them”.
(Let’s set the anthropomorphic “he” aside, which does a lot of work persuading us there’s some life in there rather than a data-processing pipeline).
I shrugged, palms up, and waited… The “ohhh” came a few seconds later as they watched the penny drop.
It’s not the first time I’ve had a conversation of this type, and probably not the last. People tend to forget that the chat prompt isn’t the only path to the LLM’s server. File access, folder permissions, or third-party integrations all open routes to that same server.
When using generative AI (or any other software for that matter), make sure you know exactly where your data is going. Don’t assume. Outside of being the right thing to do, it’s also a legal obligation.
Colin