Enrico Schaefer warns that the GPT Store lets OpenAI train on your conversations by default, with no way to know if a developer turned it off. Business users risk exposing confidential data to third-party GPTs, and developers face liability because OpenAI's terms don't automatically cover uploaded GPTs.
Chapters
As you know, the OpenAI GPT Store is about to launch. It's going to be huge, and it's going to be a lot of people who are going to be creating public GPTs for use, or basically GPT developers. And then there's going to be a lot of people using these GPTs, finding them in the GPT Store. There are a couple of things that are really important for you to know before you, number one, launch a GPT as a GPT developer. Or use any third-party GPT as well. There was incredibly high risk for both GPT developers and for users of our GPTs, which have not been disclosed by OpenAI. What OpenAI has not advised people is a serious issue for the GPT store. The first one relates to whether or not OpenAI is going to train on these input and output of user interaction with the GPT. So if you're a user of GPTs and you find a third-party GPT that you like, how do you know whether or not OpenAI is going to use your input and the output to train its model? Well, if you're inputting or uploading private, confidential, proprietary customer client information, We are obligated to ensure that OpenAI is not, is not using that input to train its model. For those of you who've seen my other videos, you know that I did a whole video on how to set up the data security settings within the web application for ChatGPT in order to make sure that they don't train on your data. And you'll click on a variety of different links in order to get to a form which allows you to opt out of OpenAI using your data to train. So it's a privacy setting where you have to find, go through the process, and then now you can keep your bookmarks and OpenAI might not train on your input, your output. GPTs are different. And so just because you opted out of data privacy or opted into data privacy, in the web app does not mean that they will not use your input or uploads to train if you use a third-party GPT. So let's just start with user risks. So let me show you what's going on here. The developer of the GPT is the only one that can check the box to opt out of training on the input, and it is not invisible to users whether or not the developer has opted out. So at the very bottom of the configuration settings for the developer GPTs, here is this checkbox, and it automatically is checked when you build the GPT. That means that you are authorizing as the developer OpenAI can use the conversation data in the GPT to train, to train its models, to improve its models. So the developer has to uncheck this box in order to keep the information input private, both the uploaded knowledge and the user input and user uploaded knowledge. So unfortunately, if you're a user, there is nothing in the user interface which tells you whether or not OpenAI is going to train on this GPT's data. There's nothing in the GPT that is available so far which advises the user whether or not the developer has unchecked this box. Well, so what does this mean? It means that as a user, you cannot use third-party GPTs that you do not trust or you do not know and have assurances that the box has been unchecked. Is going to be extremely limiting for business and professional users of AI who may want to use some of these GPTs for their own purposes, because you won't be able to use them for any confidential, proprietary, private information. You won't be able to use the GPT for work. All right, so from a user standpoint, this is a really high risk. And so what I'm telling you is you should not be using any third-party GPT through your GPT store business until OpenAI either reveals the setting, whether or not the box is unchecked, or the developer provides assurances that they've unchecked the box, as we've done with our client, you know, AssistiveGPT. As a developer, you need to know about this setting, this additional setting, and you need to advise your users if you check or uncheck this box. Okay, and there's places you can do that in the description area for your GPT, as we've done— as we've done here, GPT training unchecked. So privacy is ensured for this particular GPT that we've developed. As a developer, you need to be aware of this issue because you can be liable to users if their proprietary and confidential information gets add-in to the GPT-Train model. All right, the other issue, um, that's out there for developers and of course for users, but primarily from the developer side, is the OpenAI terms of service and use policies and privacy policies do not apply to you and your GPT that you upload to the store. Just like if you were to upload an app to the Apple Store or, uh, you know, into the Android store, you need your own terms of service and privacy policy. So I will guarantee 99.9% of these GPTs will not have provided a terms of use or a privacy policy for their GPT, which means the developer is 100% at risk, has no liability protection for users. If a user wants to sue the GPT developer they'll be able to do it, and the GPT developer will not have any of those damage limitation, class action waiver clauses, etc., that would normally be in a terms of use for an app. So how do you do that as a developer? Well, the first thing that you need to be aware of is that you do need to develop your own terms of service and terms of use. I'll put a link to this in the description, but we've got a template For you to refer to. You can't just cut and paste it; that's not going to work for you. You need to have an attorney develop it for you. But you'll see the types of issues here that you need to have in your terms of use. So this could be the terms of use for everyone that uses your GPT. You're going to some governing law jurisdiction, mandatory arbitration, class action pretrial waivers. You're going to see identification clauses for their use. There's also a damage limitation provision which says that they can only sue you for up to $100 in total damages. So that's going to be really, really important. Your intellectual property and your trademark for your GPT, any product engineering that you've done, copyright protected, you need to protect that. So I'll provide a link so you can see this and work with your attorney to develop your own terms of service. Second issue is, how do you provide the terms of service to new users or your GPT? Because OpenAI hasn't built that into the system. So there's a couple of ways you can do that. One is you can put links in your description. So you can see your GPT TOS and privacy policy found here, and I have a link to this sample, right? So you can do it that way, or you could actually build into Or prompt engineering. The requirement that you're going to first— the GPT is going to ask the user to agree to the TOS and privacy policy for the GPT. You're going to provide the user the link to your privacy policy and terms. You're going to ask them to agree. You're going to click that, do you agree? If they agree, then they'll notice the GPT. If they disagree, they cannot use the GPT. So This is the type of advanced prompt engineering that we're developing and that we're sharing with you on this channel. If you've gotten value out of this, go ahead and share it. If you know anyone that's putting GPT into GPT Store, please send them this video. You know anyone that's going to be using GPTs as a business user, send them this video. And go ahead and like this video, share this video, put anything in the comment. I will answer your questions in the comments. Till next time, I'm Enrico Schaefer. Thanks for joining the AI for the Win community.
