When OpenAI released ChatGPT to the public in late 2022, the AI service had no privacy features. That is, you could not prevent the data in your prompts from being used to train the large language models that powered ChatGPT.
It took several months and pressure from regulators for OpenAI to implement its first privacy controls in ChatGPT. And back then, it was a pretty terrible implementation. You had to either agree to allow your data to be used in training the chatbot, or disagree and have all history turned off.
Eventually, OpenAI offered ChatGPT users the option of opting out of training the model and keeping their history. That’s what I did in early November, but this key ChatGPT privacy setting wasn’t obvious. You had to hunt it down and then specifically ask OpenAI to stop using your data to train its models.
Fast-forward to late April 2024, and OpenAI made a big upgrade to ChatGPT privacy that we all nearly missed. The company now gives users the feature I always wanted. I can keep my chat history and prevent ChatGPT from using the data for training.
OpenAI revealed the new privacy features on X, without a formal announcement. That’s why I and so many others overlooked it, especially since it came right after OpenAI’s big memory upgrade for ChatGPT.
The company says the new privacy feature applies to both ChatGPT Free and Plus users:
Now, you can access your chat history regardless of whether you’re opted into training for model improvement. If you’ve previously opted out, your choice will remain. Available on the web today and mobile soon.
You’ll have to go to your ChatGPT account to take advantage of it. Tap on your profile picture, then choose Settings and then Data Controls. You’ll find a toggle called Improve the model for everyone that you can switch on and off.
As you can see in the image above, mine is turned off. I discovered a way to opt out of ChatGPT training months ago and used it to shield my data from OpenAI. This new update makes things much easier.
I’ll also note that OpenAI made a big change to the name of the privacy setting. Previously, it was called Chat History & Training, as seen below.
But, of course, Improve the model for everyone sounds much better. It’s a message to ChatGPT users to keep feeding OpenAI with data for ChatGPT. There’s nothing wrong with that.
ChatGPT and other chatbots can’t improve without data to train with. I’d love to help “improve the model for everyone,” but not with all the data in my prompts. It would be great if OpenAI could give users more granular control. I’d select certain chats to share with OpenAI for training ChatGPT while preventing the ones that might contain sensitive information from being used to improve the models.
Finally, I’ll also mention the Temporary Chat feature available in ChatGPT. Announced last week as part of the big memory upgrade, Temporary Chats will not appear in your history and will not be used for ChatGPT memory or training the chatbot.
If you care about privacy in the age of AI, you’ll almost certainly want to take advantage of these new features.