Small thing happened this week that I've been low-key annoyed about for months: OpenAI finally let you turn off chat history in ChatGPT. It landed last Tuesday, and if you dig into Settings there's now a toggle called "Chat History & Training" that, when you flip it off, stops that conversation from being used to train future models and keeps it out of your sidebar. Conversations started with it off get deleted from their systems after 30 days, per their own explanation, kept around only long enough to check for abuse.
I know this sounds like a small UI change to complain about but hear me out. Since December I've had a running mental tally of things I've pasted into that text box that I probably shouldn't have. A chunk of a client's config file. A paragraph from an internal doc with names still in it. A half-finished cover letter with my actual home address on it because I was too lazy to redact. None of it was catastrophic, but every time I hit enter there was this little voice going "so this is just, what, sitting on a server in San Francisco getting read by a contractor labeling training data at some point." Turns out that voice was right, more or less, and now there's finally a switch.
The timing is not a coincidence. Italy's data protection authority, the Garante, banned ChatGPT back at the end of March over exactly this kind of concern, no age verification, no clear legal basis for hoovering up personal data to train the model, no way to correct or delete what it had already learned about you. OpenAI went dark in Italy for about a month while they negotiated. On Friday the Garante said okay, you can come back, after OpenAI agreed to add things like an opt-out for training, better disclosure in the privacy policy, and some kind of age-gating. The history toggle is clearly part of that package, rolled out globally instead of just for one country, which is honestly the sane way to do it.
I'll say the plainly good part first: this should have existed at launch. Not eventually, not after a European regulator forced the issue, at launch. A product that logs everything you type and treats it as fair game for model training by default, with the burden on the user to find a toggle five months later, has its priorities backwards. I don't think OpenAI is uniquely evil here, this is just how the whole industry defaults, but I'm allowed to be annoyed that it took a government intervention to get a checkbox.
That said, I flipped it on last night for basically everything and it does what it says. No history saved, and there's a note that says these chats won't be used to improve their models. I lose the ability to go back and find that one prompt from three weeks ago where I finally got it to write a decent regex for stripping trailing whitespace without also eating the newline, which, fine, tradeoff accepted. If I want to keep something now I just copy the good output into a plain text file like it's 2004.
One thing worth knowing if you're going to do this yourself: the toggle is per-device right now, or at least it behaved that way for me, I turned it off on my laptop and it was still on when I opened ChatGPT on my phone later that day. Had to go flip it in both places. Small annoyance, but if you're doing this specifically because you're worried about sensitive stuff leaking, don't assume one flip covers you everywhere you're logged in.
There's also apparently a "ChatGPT Business" version in the works aimed at companies, with data controls baked in by default instead of bolted on, which tells you where OpenAI thinks the real money is going to be. Makes sense. Nobody's legal department is going to sign off on employees pasting customer data into a chatbot that trains on it by default, and OpenAI clearly knows that.
Anyway. If you use ChatGPT for anything even mildly sensitive and you haven't gone and found this setting yet, go do it now. It's under the three dots next to your name, then Settings, then Data Controls. Took me longer to find than it should have, for what it's worth.