HN Explorer

Tell HN: ChatGPT is the biggest privacy disaster ever

15 points by I_am_tiberius ↗ HN
A lot of people share personal information with ChatGPT. Most people know this. But many might not know that what you write is by default used to train future versions of ChatGPT.

It's more concerning because the setting that lets ChatGPT use your chats for training is turned on automatically. This setting is saved only on your device, not OpenAIs servers. So, it's difficult for you to proof that you disabled the flag (if OpenAI is audited, they can say they never received prompt queries that asked for respecting the users' privacy).

There's also a new feature where you can give personal instructions with each message. These instructions can include personal details. The setting for keeping your prompts private ("Chat history & training") doesn't affect these instructions. To turn off this feature, you have to email OpenAI, but finding the right email address is hard, and OpenAI is known not to respond quickly.

I wanted to warn people about these privacy issues. I think the European Union might do something about it, but it could take a year or more for them to realize the degree of harm that is currently being caused. It looks like Sam Altman, the CEO of OpenAI, is pushing the boundaries. This situation reminds me of what Marc Zuckerberg did, but it seems much worse.

Also, when you turn off the "Chat history & training" setting, the chat sometimes clears itself. This happens quite a lot. I'm not sure if it's intentional or a mistake. Since it's been happening for over a month and affects my colleagues too, I think it might be done on purpose in order to annoy users who prioritize privacy.

FYI: It seems that OpenAIs models still include training data 1:1 (see https://news.ycombinator.com/item?id=38499575).

11 comments

[ 0.21 ms ] story [ 71.9 ms ] thread
eimrine
Service-as-Software-Substitution. It should be taught in school, but Microsoft and Apple keeps pushing their narrative to the kids. And the narrative is pretty opposite to values of freedom, privacy and reverse engineering.
kkzz99
You can disable that your content is used for training[1].

[1]: https://help.openai.com/en/articles/5722486-how-your-data-is...

zxt_tzx
Related link: https://help.openai.com/en/articles/7730893-data-controls-fa...

> This setting does not sync across browsers or devices. You will have to disable chat & history on each device/browser.

I wonder whether this is a dark pattern or if there is a good reason for it.

I_am_tiberius
This is what I meant by:

> This setting is saved only on your device, not OpenAIs servers. So, it's difficult for you to proof that you disabled the flag (if OpenAI is audited, they can say they never received prompt queries that asked for respecting the users' privacy).

zxt_tzx
Ah I see. You're right, the more I think about it, the more I'm convinced it's an intentional dark pattern.
I_am_tiberius
What's also interesting is that my post was trending in the Ask section on hackernews and suddenly disappeared from there.
danielbln
It might be a conspiracy, or it might just be a lot of users going "well, duh?" or disagree with your notion. At least you can control what you put into the system, where companies like Google vacuum every behaviour from your emails, browser history, location data etc. That doesn't make ChatGPT any less problematic in terms of privacy, but I certainly don't think it's the "biggest privacy disaster ever".
I_am_tiberius
Please read my post description. Also, have you checked the "this form" link of the page you linked?
nulcow
To be fair, I'm pretty sure they tell you that upfront, alongside warning you not to send any personal information in conversations.
Jerrrry
The AGI won't think we are apart from it; why should it - and why should we doubt it? - when it knows more about and between us than we do.