How ChatGPT changes the goals of OpenAI’s Terms of Service

During ContractsCon 2023, Foster Sayers joined Mike Whelan to discuss OpenAI’s Terms of Service and their application to ChatGPT.

According to Sayers, the terms of service may have applied well to the users of OpenAI’s API connections, but perhaps not everyday consumers. Now that ChatGPT is explicitly targeting consumers, should their terms change?

Watch as Sayers covers the contract’s odd rules, including:

• An unrealistic age restriction,
• A rule against using OpenAI’s name combined with disclosure rules,
• Confidentiality terms, and
• Unlikely arbitration op-outs.

 

 

Episode Links

The Contract:

https://web.archive.org/web/20230228014736/https://openai.com/policies/terms-of-use

Guest’s Links: LinkedIn | ContractMVP | Pramata

Interview Transcript

Mike Whelan In this special live recording from Contracts Con 2023, Foster Sayers breaks down OpenAI’s terms of service and how they apply to ChatGPT. So let’s tear it down. We’re going to beat up the ChatGPT. Well, we’ll talk about this: the OpenAI terms of service. We’re doing it with my friend Foster Sayers. How are you today, Foster?

Foster Sayers I’m doing great.

Mike Whelan All right. So this document, ChatGPT has been in the news. My mom knows what it is, which is crazy. So we’ve all heard about it. We know it. But people sort of jumped on it. They started using it without really knowing what they were doing. And so we want to talk about that experience and what happened with this ship with ChatGPT coming out. Before we dig into it specifically, Foster, tell us about you. What’s your background? What brings you to documents like this?

Foster Sayers So I am general counsel and Chief Evangelist for Pramata, a contract management software company, and I joined them after almost 15 years as a commercial contracts attorney. And now I also teach webinars on contracts.

Mike Whelan Right. Well, you are our kinda nerd. So that’s what we’re going to do here is nerd out a little bit about the terms. Laura is going to be pushing buttons back there, so blame her if it doesn’t happen as fast as you want, but they’re going to be up on the board here. And also reminder to check your email if you don’t have the document in front of you, if you’re watching virtually, check your email right now. There’s a link there for you to be able to get access to thing to read along as we go. So let’s start at the beginning. If you look at the very top and before we dig into it, tell us a little bit about what OpenAi is. What’s the relationship between OpenAI and ChatGPT?

Foster Sayers So as it says, these terms apply when you use the products or services of OpenAI LLC. And so ChatGPT being one of their products or services is subject to these terms. And ChatGPT is an interface, right where you can have a conversation with it about a topic and it can range from asking it to review computer code to lots of other things.

Mike Whelan And you’re going to see a theme through this is OpenAI was for super nerds, ChatGPT is for my mom. What’s the contractual relationship change that happens between those two things? So let’s start at the beginning. Under registration and access, it says you must be 18 years or older and able to form a binding contract with OpenAI to use the services. What do you think about this as an immediate block to my kids getting on and using ChatGPT.

Foster Sayers It could be stated as a clear prohibition such as you shall not use the services if you are not 18 years of older or as it’s framed. It’s almost like this person just has an ax to grind with like, say, Disney’s The Little Mermaid and the fact that you could have destroyed the whole plot premise by saying, well, she’s not the age majority, she lacks the capacity to contract. The contract is voided. So I think what they’re saying is, yeah, you can’t agree to a contract with OpenAI if you’re not 18 which is true but they could do a much better job of very, very clearly prohibiting the use by minors.

Mike Whelan So ChatGPT in this case is Ursula and is saying you have to be above 18 if we scroll down to the different restrictions that they put on this, this is in 2C. This is super interesting because if you happen to use ChatGPT – I use ChatGPT whenever I can get access to it as a brainstorming tool. So if I have to write something, I do a lot of writing, but so I already know in my head something about the thing that I’m going to write. I use it to unlock writer’s block. I’ll go on there and I’ll say, Hey, give me five reasons that somebody should watch the Contract Teardown Show, and it’ll give me five reasons. And they’re usually bad and clumsy, but I’ll go in and I’ll clean it up. Well, then the question is we’ve got output, right? I put in inputs, which was the question, and now we’ve got the output. What am I going to do with that output? If you look in 2C, they put a restriction on the use of OpenAI including ChatGPT under five that says, you may or may not use the services if you represent that the output from the services was human-generated when it was not. Tell me what that means. If I’m writing a blog post, my name is at the top. What’s my relationship with ChatGPT under this language?

Foster Sayers So the way this restriction reads you may not represent that was you that wrote it if you did not. But. There’s a catch. Because if we look later in their terms in 9B.

Mike Whelan Laura, let’s scroll down to nine. If you look at B, it says use of brands. You may not use OpenAI’s or any of its affiliates – names, logos, trademarks without prior written consent, so don’t tell people you wrote it, but don’t tell it.

Foster Sayers Don’t also tell people.

Mike Whelan We wrote it.

Foster Sayers Unless you get their permission in advance, which I wonder if there’s anyone at their business that knows that they were set up for this paradigm where they’re going to need to approve every user’s use of attribution because you can’t attribute it to yourself, but you can’t attribute it to them either without written permission.

Mike Whelan Yeah. Feels like no matter how my mom uses this thing, she’s going to get in trouble. Let’s jump down to five and confidentiality. I’m seeing a lot of talk about this, about what happens with the data you’ve created. So you might be inputting actual code, you might be going in and creating company secrets for all you know, or you could be putting all kinds of information in there that might not be very good for public consumption. Let’s go to confidentiality. The sending starts confidential information means nonpublic information that OpenAI or its affiliates or third parties designate as confidential, or should reasonably be considered confidential under the circumstances, including software specifications and other nonpublic business information. What do you think about that definition?

Foster Sayers I think that it excludes anything that you put in to the ChatGPT to have a conversation, because fundamentally the definition of confidential information does not extend to any information that you share, meaning that, you can expect that it will be maintained in confidence. And in fact, if you read the FAQ for ChatGPT, they do state that humans can read the conversations, which they do for QA and training purposes.

Mike Whelan Yeah. There was a conversation around this about computers are racist, right? Like these computers, you know, they’re taken from the Internet. They’re taken from the craziness that is all of us. And so a lot of these AIs, when they go in and they create a lot of it, it’s stuff that’s ugly. Right. It’s a very simplistic representation of the Internet. And the Internet is freaking crazy. And so some of the stuff that, you know, they have to hire humans to go in and make sure there’s some filtering involved, there’s some cleaning up, there’s some research involved for how they’re going to make the tool better. So there are humans reading it. I’m looking down at seven now under indemnification. It says that you’ll defend, indemnify and hold harmless us, our affiliates, our personnel from going against any claims losses, expenses including attorney’s fees arising from your use of the services. Do you think that this kind of indemnification matches that very human touch involvement?

Foster Sayers Well, what’s interesting is that there’s no discussion about how that, you know, defensive claims going to be controlled. So if you have my mom indemnifying you and she does something really crazy that harms a lot of people, is that notification going to matter much? And do you want her son defending your case because she indemnified you? It’s really one of those things where you as the user, though, need to be aware that you are completely responsible for whatever – they’re not they’re not taking any ownership of any consequence for any output, whether it be racist or whatever.

Mike Whelan Which is nice. And as you see in documents like this, often there’s a dispute resolution section under A – there’s mandatory arbitration. It says you have to do arbitration. Right. And this is common when you have these platform businesses. There’s just too many disputes for it to be otherwise. But they put in here that you have the right to opt out of these arbitration terms and future changes to these arbitration terms by emailing arbitration-opt-out@openAI.com within 30 days of agreeing to these arbitration terms or the relevant changes. What do you think?

Foster Sayers I think that you should check the last time when you first started using ChatGPT, you maybe sent an email because if you have to fully indemnify them, then you probably want a jury trial at least.

Mike Whelan Yeah. Needless to say, I did not send arbitration-Opt-out.

Foster Sayers We didn’t know that you needed to do that.

Mike Whelan You know, as I look at this and obviously there’s a lot of conversation about ChatGPT, and I’m sure that you all are getting questions about what to do with these kinds of tools in general. I’m noticing something in the drafting exercise that has, my what’s left of my hair on end. There was a change. Like, we talked about this OpenAI when it was originally created, we did a previous episode with Justin about OpenAI. It was a big company people were using it, but it was mostly super nerds. It was mostly people using the API for creating stuff. Well, now my mom is using it and her relationship with this tool is very different than a technician going in and using OpenAI, yet they haven’t drafted anything new. When you think about a document like this, when you get this sort of consumer focused explosion of use overnight, like you saw this with a tool like Facebook, they opened up to non-college addresses and it exploded. When you have a fundamental change in the business alignment like that. How does that change the contracting exercise, how do you draft terms?

Foster Sayers Well, it should be changed to have a consumer focus instead of a business focus. And the way that you put terms in front of consumers is very different, primarily because a lot of times there is personally identifying information, other types of information that’s regulated that you want to be sure that you set expectations with your consumers about what they can input, what they shouldn’t, and you know what their rights are and are not. And then if you don’t do a good job of gating those consumers the way that regulations are proliferating around consumer data, you can put yourself in a bad position.

Mike Whelan Yeah. The other thing I was thinking about it as remembering when I was in high school, I was supposed to read a fellow in this class and I didn’t because there was a movie, there wasn’t a movie of Othello yet, but there was an operetta called Otello, which had Placido Domingo having painted his face black, which do not do that. And I watch this operetta and I go through it and I’m man, I’m confident and I go take the test and and I tell. The ending of Otello is totally different than the ending of Othello. And my teacher wrote at the bottom, Are you messing with me? Because I gave totally wrong answers. To me, that is ChatGPT, right? Like it has the confidence of my 16 year old son who’s an idiot and just says things as if they’re very true. Do you feel like if a consumer, especially my idiot son, is going in and using this thing, that this document properly tells them, Hey, here are the risks, we’re going to be wrong half the time. If you go out and use this thing, you know, does this properly contract for that actual use?

Foster Sayers Well, proper contract is a subjective term. What I would say is it does a poor job of setting expectations with the users for how that tool is actually going to deliver information to them. And that’s a problem, right? Because some people will rely on it for certain things, not being aware of that, which could lead to poor outcomes. But additionally, I think it’s a situation where because consumers are just in the mode of click to agree, that, you know, they’re going to miss that. Oh, I should have opted out of arbitration before I started writing my entire blog with all this stuff and wasn’t paying attention that it snuck in this thing in a paragraph and now we’ve got problems.

Mike Whelan Well, Foster, I appreciate you joining us. For those of you who are watching, we do this every week. If you want to be a guest on the Contract Teardown Show, please email us at community@lawinsider.com. Foster for you, if people want to learn more about what you do, connect to the online. What’s the best place to get you?

Foster Sayers You can find me on LinkedIn and you can also find me at my Contract MVP channel on YouTube and my company at Pramata.com.

Mike Whelan We’ll make sure to have links to that and to this document over at Law Insider.com/resources. We’ll see you all the next episode. Thanks for joining us in Miami. Thank you.

Foster Sayers Thanks.