I got these questions during the events where I recently spoke about LLMs and security.
Yes, an LLM is "processing" your data when you communicate with it.
But does this "processing" mean your data is chopped into tiny, unrecognisable fragments, impossible to reconstruct?
Like trying to turn chicken nuggets back into a chicken.
Or does it mean everything is stored and indexed, with the ability to retrieve any detail at any given moment? Are they listening? Watching? Memorising?
My answer is: I don’t know.
What I do know is that governments have historically gone to extreme lengths to access information.
So when this question comes up, I always share the story about Crypto AG.
Crypto AG was a Swiss encryption company. It sold highly encrypted communication systems to governments and it did so for decades. Crypto AG sold equipment to more than 120 countries.
➡️ Their systems had a backdoor.
➡️ The CIA and German intelligence secretly owned the company.
➡️ They listened to everything, for decades.
➡️ Crypto AG was operational until 2018 (!)
If something has value, someone will try to get to it. So we shouldn’t be naïve about what we share and how freely we plug our systems into external models.
I’m not saying this is Crypto AG all over again. But it’s OK to be a little paranoid. It’s your data.