I’ve been using ChatGPT-4o Mini on my IPhone (Mobile) without an account for the past few days and have been asking it some pretty personal prompts and questions, relating to a personal situation I am experiencing right now. ChatGPT Mini replied back to one of my prompts saying “I’m sorry, but I can’t assist with that”, but that’s it. Does this mean I have been flagged? Will I get in trouble for what I asked?
I have never had any orange or red flags or notices come up from my prompts about potentially violating usage and policies in GPT 4o Mini. If I said something violating, would a notice show up or would that not happen in the Mini ChatGPT? How do I know if I have been flagged or not, and will the developers review my prompts/conversation?
I have been freaking out over the past few days wondering if any of my prompts have been flagged or not.
Any help explaining these questions is greatly appreciated!
First of all, you should be expecting anything that you enter on ChatGPT to be read by someone at some point. It probably won’t happen nor would the person probably care, however it’s good to treat sensitive information the way it should be treated - with care.
If you don’t want anyone to find out any information about you or your situation, do not use a public service.
That being said: You probably won’t get in trouble as it takes multiple offenses for you to really get in trouble - if even at all. At most you would probably get banned from using the platform, however I think this is pretty unrealistic. Do check on what OpenAI says about this in their guidelines and FAQ.
Everyone gets their prompt flagged at some point and even multiple times, I haven’t been banned for this even though some prompts of mine have been flagged.
All in all, try to avoid sharing private information with ChatGPT or any public app as internals can theoretically read your messages.
Thank you for this. I think ChatGPT deletes prompts/conversations after 30 days if I’m not mistaken, unless prompts are kept for “safety/abuse” purposes, which I am worried about. I did not have it opted-in for training the model at all, and was only using the “Try it now” feature online.
With regards to the other questions I asked, I can’t find anything about them in the FAQ, forums, or the usage policies. I’m wondering if the reply from the AI “sorry, but I can’t assist with that” is counted as a flag? If it’s not, I’m also wondering if GPT Mini 4o does in fact notify you in red or orange text about policy violations or being flagged, or if that is only available in the previous versions?
I have also once had it not reply back to me at all, and just show up as a dot beside the prompt, but I re-asked the prompt and it answered.
Hopefully you or someone else can assist with this.
This is generally just a soft refusal, you’ll likely be able to get around it with some clever reformulation.
That said, you should absolutely heed what @j.wischnat said. Anything you send and generate in the app can be viewed by openai employees and/or contractors and may or may not be used for training future models. What they do or don’t view may be completely unrelated to whether that content was flagged or not.
Even the API is subject to review and inspection unless you have a very specific agreement with the service provider.
Your privacy is absolutely not guaranteed here
That said, purely statistically, the probability of your chats getting inspected by a human are likely pretty low. But not zero.
It’s kind of sad that this technology isn’t really accessible to people who might really benefit from it for therapeutic reasons, but the only way to really use it for that is to host the model yourself
This is most likely correct. If you’re afraight that someone will read your messages, it is still best practice to not make the information public in the first place - even if it gets deleted after some time. Treat that information as if it will be available to everyone, forever.
There is no guarantee your information will stay private even if a company says so - anyone could get hacked and leaked - even OpenAI.
I’m not sure on this, hopefully someone can answer this - however I think it’s just a moderation happening in the background and just refusing to answer the prompt instead of an actual flag that will have consequences.
This is pretty normal and happens sometimes - no issue at all.
If you want to have complete control over your information and make sure it never gets read by anyone but you - do as @Diet said and host a model locally - there are some really solid options out there by now.
Thank you for your help, and @Diet too. I’ve actually spent the past couple of hours reading through the data retention policy for general temporary chat users, who use GPT4-0 mini for example, and have figured it likely goes as follows:
Your prompts and other data is automatically defaulted to not opt-in for training the models & services.
A: IF you choose to opt-in, your data will be kept for as long as needed to train the models, but it will be de-identified and aggregated.
B: IF you do not opt-in, your prompts, files, and conversations will be deleted/purged from the system in approximately 30 days, unless held for security/abuse purposes.
Does that sound right to you? Thank goodness I never opted-in with the information I gave LOL.
I strongly encourage you to still apply my advice. I don’t want to seem all-knowing or wise - but this is a thing everyone should excercise.
If you don’t want anyone to see your data, do not publish it in the first place.
Imagine OpenAI got hacked - do their policies protect your data from being leaked by a hacker with malicious intent? → Definitely not!
Personally identifiable information has to be treated with care - especially in EU countries.
As OpenAI allows users from the EU they HAVE to respect and abide the GDPR law.
This applies to any legitimate company.
However, this does not guarantee the preventation of slip-ups and leakage of personally identifiable information.
Leaks happen all the time - please, be safe and don’t share any personal info with ChatGPT that you don’t want leaked.
Hi again @j.wischnat , I made another post and would very much appreciate it if you could help answer, as I’m very confused if third-party websites can see and keep our prompt data. I just want to know if my data in those two websites will be used for model training, or if they will be deleted after 30 days as well?
Thank goodness I never opted-in with the information I gave LOL.
