The Fine Print of AI Terms of Service: What Most Businesses Miss

AI platforms are helpful for content creation, coding, product development, and numerous other tasks. However, many people don't actually read the Terms of Service for common AI platforms, and just click "Accept".

This exposes your business to legal and operational risks, as there are clauses that could affect important issues including IP ownership, confidentiality and privacy, business data being used for model retraining, usage restrictions, and more.

This article will cover what the major AI platforms are in 2025, what Terms of Service are and what they cover, and which clauses you should watch out for in any AI Terms of Service, with practical examples from the major platforms. Then, we'll cover some tips for how to protect your business when signing up to use a model.

Let's get started.

What are the Major AI Platforms in 2025?

The major AI platforms in 2025 include Google Gemini, ChatGPT from Open AI, Claude from Anthropic, Llama from Meta, and Grok from xAI.

Smaller AI platforms are out there too, but most are not widely used by businesses. Others, like the DeepSeek open source AI model are usually not used outside of China because of privacy concerns.

Each of these AI platforms has its own Terms of Service (ToS), which is important to read through before you begin using it. This is particularly the case because some of these ToS differ significantly from each other, and you may be exposing yourself to legal or other risks if you don't read them carefully.

What are Terms of Service?

Terms of Service (ToS) is another name for Terms of Use (ToU), or Terms and Conditions (Ts&Cs). These are legal contracts between a company, such as an AI platform company, and its users.

ToS cover how the service, in this case a large language model (LLM) or other generative AI platform, can be used. They set out what rights and responsibilities apply to the user and the company, how intellectual property is dealt with, what warranties or limitations of liability apply, and other more general terms such as governing jurisdiction for the contract.

With AI ToS, some of these terms are particularly important, such as who owns generated output, whether user input data can be used for model retraining, and how confidential or private information is handled, among other things.

Let's take a look at some of those important terms now.

Which Types of AI Terms are Important to Note?

Some terms in ToS are more important to take note of than others. While you should always read the whole contract, many businesses and people simply skip through and click "I Agree", without actually checking the terms.

This means you can easily miss important clauses that can significantly affect your business when using a generative AI platform.

The most important clauses you need to consider and read carefully include:

• IP ownership
• Confidentiality
• Retraining rights
• Limitation of liability
• Usage restrictions
• Updates

Each of these has particular aspects that you need to note, to make sure your business is not signing up for something that could affect your IP, confidential data, data use, or otherwise.

This article will go through the Terms of Service for some of the major platforms for each of these clauses, so you can see what you need to look out for.

IP Ownership

Intellectual property (IP) ownership rules for each platform can vary widely. It's important to check these clauses to determine who owns the output of an AI model, and what happens to the IP ownership for input. In addition, as these are relatively new technologies, many issues are still being litigated. This means that rules and clauses could change in the future.

For example, several copyright lawsuits have been combined into one case against OpenAI, from many publishers, writers, and other creators who say their work has been used without permission. Depending on how this case and others like it play out, it could affect how liability is determined when using AI tools.

If you are using AI tools, you'll need to be careful that they are not using copyrighted or other protected content. Or, if the model generates something that infringes upon someone else's work, you could be liable for breach. These issues will become clearer as litigation develops, but it's an important issue to keep in mind.

Aside from these IP issues with the use of copyrighted content, you also need to consider how your business IP relates to the IP clause of the AI model.

Check whether the clause gives you full ownership over anything generated by the AI model, or whether the AI company claims a license over it. You may also need to consider whether any content that the model produces can be commercialised.

For the model xAI, made by Grok, all input and output are classified as "User Content". Their Terms of Service state: "To the extent permitted by applicable law, and as between you and xAI, you retain your ownership rights to the User Content". This means that IP ownership is still kept by you.

However, as you can see in the terms of Grok below, an "irrevocable, perpetual, transferable, sublicensable, royalty-free, and worldwide" license is granted to xAI, to "use, copy, store, modify, distribute, reproduce, publish, display in public forums, list information regarding, make derivative works of, and aggregate your User Content and derivative works thereof for any purpose."

You can also see that xAI specifies it can use the User Content to "maintain and provide the Service", to "improve our products and the Service".

In addition, xAI can use any of your input for analytics, marketing, product improvement, and more. Even though you continue to own your input into the AI model, xAI has effectively free reign with how it uses it.

You can also see in the clause below how xAI manages its own intellectual property rights, and what it does with usage data. Usage data is "information about your computers, mobile devices, systems, and software", and is owned solely by xAI.

For Google Gemini, there are two main sets of terms that you need to consider. These are the Gemini API Additional Terms of Service and the Google APIs Terms of Service.

The Gemini API Additional Terms of Service explain who owns the IP for generated content. It states that Google won't claim ownership of any generated content.

The Google APIs Terms of Service set out how the IP for submitted content is dealt with. It sets out that Google does not acquire ownership in content you submit to APIs such as Google Gemini.

However, like xAI, you grant a license to Google to use the content submitted.

In the Meta Terms of Service, no license is claimed. Instead, Meta's ToS states that you continue to own all rights, title, and interests in both input and output. Meta does retain content for 30 days to provide the services, and for compliance with laws.

OpenAI's Terms of Service state that your ownership is retained in any input into the model, and that you also own the output. OpenAI also does not claim a license over any inputs or outputs.

You can see that these terms vary quite significantly in terms of IP ownership and licensing, with some AI companies claiming no license, and others a very extensive one.

Now let's take a look at confidentiality.

Confidentiality

Confidentiality and privacy clauses relate to how AI models deal with confidential or private data.

If you are a business using an AI model for your work purposes, you may be submitting confidential information or trade secrets into the model. First, you should avoid this if possible. However, many Terms of Service also cover how privacy and confidential information are dealt with. Others do not.

The submission of confidential or sensitive data becomes complex for a number of reasons, not least is that you could be vulnerable to exposure.

In addition, when issues like the OpenAI litigation arise, courts may decide to retain data for legal purposes. In the OpenAI case, the court ordered the company to keep all ChatGPT conversations indefinitely, including explicitly deleted conversations.

These court orders or consequences of litigation can not be easily predicted, and are often not reflected in the Terms of Service. Some AI companies refer to their Privacy Policies, which may state that data will be retained for legal purposes.

For example, on the Gemini Apps Privacy Notice, which supplements Google's Privacy Policy, it states that data can be retained for a number of reasons.

These issues with privacy, confidentiality and deletion, should be considered carefully when using generative AI, and you need to read both the Terms and the Privacy Policy carefully.

Now let's take a look at some Terms of Service and how they cover this issue.

The Terms of Service from xAI below do not have any information about confidentiality. However, the ToS does have a section on privacy and security, which simply refers to the Privacy Policy. It also explains that data will be transferred to the US for processing.

In the section from Google below, you can see that confidential information is covered, however, primarily relating to Google's confidential information. The clause sets out that if you, through the use of Google APIs (including Gemini), find out any confidential information about Google, that you should not disclose it.

In the example from Meta below, confidentiality is referred to for both parties. This means both Meta, and your business. It explicitly states that "Confidential Information includes your Content".

You can also see that it sets out that "each Recipient (i) will use reasonable standards to protect Confidential Information, and (ii) will not disclose the Confidential Information of the other party to any person or entity other than its officers, employees, and consultants who need access to such Confidential Information".

As you can see, confidential information is not explicitly covered in many AI ToS, although it can be very important. Take care that you don't submit confidential information to AI models, and if you do, make sure the ToS includes terms to protect it.

Retraining Rights

Retraining rights are the rights that AI companies claim over using your input to continue to train their models. Before you submit information to AI models, make sure that you know what retraining rights AI companies are claiming over data that you submit.

In the example from xAI below, you can see that you can only opt out of model training if you are logged in.

If you are not logged in, on the other hand, "you grant us [xAI] full rights to use any data you provide to or obtain from our Service for product development and model training purposes."

In the example from Google below you can also see a similar approach. If you are using Google's unpaid service, Google "uses the content you submit to the Services and any generated responses to provide, improve, and develop Google products and services and machine learning technologies."

However, "Google only uses content that you import or upload to our model tuning feature for that express purpose." This means that some data you submit will be used for improving machine learning technologies, but will not be used for model tuning. Other data you submit specifically for model tuning will be used for that purpose.

You can also see that Google explicitly states that you should "not submit sensitive, confidential, or personal information to the Unpaid Services", as this article mentioned in the previous section.

You can also see in the section below that when you use paid services, Google does not use your data to improve products.

In the example from Meta below, your content (i.e. your input) is not directly used for model training. However, data in aggregate is used for testing, supporting, developing, and training Meta services, such as the AI model.

Anthropic, which makes the AI model Claude, also explicitly states in its Terms of Service that Materials, which include input and output, are used for model training unless you opt out.

Anthropic's ToS also explicitly states that even if you opt out, some of your input and output will be used for model training, such as when you provide direct feedback about any of it, or if there is a safety issue.

In the example from OpenAI below, a similar approach is taken. You can see that content is used to develop and improve services (including ChatGPT), and if you don't want your content to be used to train models, you can opt out.

Now let's take a look at some other clauses.

Limitation of Liability

Another clause you need to take note of is the limitation of liability clause.

If your business suffers a loss as a result of an AI model, you may not be able to claim any damages. This depends, to a large extent, on what jurisdiction you are in. This is because in some countries, limitation of liability clauses are valid. In others, they are invalid, or only certain liabilities can be limited.

In this example from xAI, you can see liability is limited in most cases. However, at the bottom you can see that "Some countries and states do not allow the disclaimer of certain warranties or the limitation of certain damages, so some or all of the terms above may not apply to you, and you may have additional rights".

This means it is important to check your jurisdiction and how limitations of liability are dealt with.

In the example from Google below, you can see a very similar limitation of liability clause. Most of these clauses are standard, but it's still important to check them and find out what they state, and to determine whether the clause is enforceable in your jurisdiction.

Here's one more example, from Meta below:

Note that the clause states "to the fullest extent permitted by law". This means that some jurisdictions and laws limit how much liability can be limited, as mentioned above

Usage restrictions

You also need to take note of any usage restrictions in the Terms of Service from an AI provider.

This is because if you use a model for a use that is restricted, your account could be banned and you could lose access to the model. In addition, if you have paid for the service and breached the usage restrictions (resulting in termination), you will likely not receive a refund.

Google has set out use restrictions such as region restrictions, as shown below. There are also restrictions on what you can use Google Gemini for, such as developing competing models.

You can also see that you are not allowed to attempt to reverse engineer Gemini, or bypass any protective measures for Gemini. There are sometimes other specific rules, such as for Gemini that you are not allowed to use it for medical or clinical uses.

In the example from Meta below, the ToS also outlines that you are not allowed to use the AI platform to hide your identity, to reverse engineer the platform, or to submit, for example, children's data.

You can also see similar restrictions from Anthropic for Clause below.

You are not allowed to use the model to violate laws, develop competing products, reverse engineer the AI model, scrape, crawl, or harvest data, or to obtain unauthorised access to any system.

Each Terms of Service has its own list of restrictions on use. These can vary widely.

For example, for OpenAI below, you are not allowed to use ChatGPT to violate anyone's rights, modify, copy or lease the services, reverse engineer the model, programmatically extract output, interfere with the services, or create competing models.

Note that OpenAI also prohibits using ChatGPT and representing to others that the output was human generated, when it was not.

Now let's take a look at updates.

Updates

Finally, you need to check the updates clause in any Terms of Service.

Terms can be changed at any time, and if you continue using the model this could be considered by the AI company to be an acceptance of the new terms.

This approach (continued use being deemed as acceptance) is almost always not legally enforceable, however. This is called a "browsewrap" agreement. Instead, "clickwrap" agreements are the appropriate standard for being bound by terms, in which you explicitly agree to the new terms. Some courts have found browsewrap agreements enforceable on rare occasions though, so it's important to be careful.

In this example from xAI below you can see that the clause states "Your continued use of the Service after any change to these Terms constitutes your acceptance of the new Terms of Service."

In the example from Google below you can see very similar wording. It states: "Your continued use of the API constitutes your acceptance of the modified Terms."

You can see this repeated again in the clause from Meta: "Your continued access or use of the Services after the modifications have become effective will be deemed your acceptance of the modified Terms".

However, these ToS also state that "If we make changes that are material, we will use reasonable efforts to attempt to notify you, such as by email and/or placing a prominent notice in your Managed Account."
This means that you will receive notice of modifications in various ways and the changes will be more likely to be enforceable, because they will have been brought to your attention.

You can also see in the example from Anthropic below, that it states "If you continue to access the Services after we post the updated Terms on Anthropic's website or otherwise give you notice of Terms changes, then you agree to the updated Terms."

This implies that they will give you notice of any updates in some way. Whether this notice is effective, will be dependent on the circumstances.

Finally, let's take a look at how to protect your business while using these AI models, so that these Terms and Conditions don't catch you unaware.

How to Use AI While Protecting Your Business

There are a few things you can do to help protect your business when you are using AI. These can help you to get better terms to begin with, or to avoid signing up to tricky Terms of Service from the outset.

These include:

1. Create internal AI use policies: First, make sure your business has an internal AI use policy. This makes sure everyone is aware of whether they should be using AI models with business information, and if so, how to do it safely.
2. Negotiate terms if your business is big enough to do so: If your business is very large, you may be able to negotiate terms with an AI model supplier. For small businesses this is unlikely to be an option, so make sure you review the Terms of Service for any model you do use.
3. Upgrade plans: If possible and financially viable, upgrade plans. Often, paid plans have better conditions than free plans, particularly when it comes to data use and model retraining.
4. Safeguard your data: For any model you do use, safeguard any confidential and private data. You can establish rules with your employees about what data is suitable to be submitted to these models. This can be incorporated into your AI use policy, and should be frequently reiterated to employees.
5. Review terms regularly. Finally, review terms regularly. Terms of Service can change frequently and may change in significant ways. Make sure you check whether any modifications have been made, and consider whether you still want to continue using that model under the new terms.

Summary

AI platforms are useful tools for many businesses, but there are a number of legal and operational concerns you need to watch out for. When signing up, do spend time reading through the Terms of Service, and take particular care when reading sections about IP ownership, confidentiality, retraining rights, limitations of liability, usage restrictions, and updates.

Create an AI usage policy for your business, and take care not to submit any sensitive information to the model. Pay for upgrades or negotiate terms if possible, and for any AI model you do sign up for, review the terms regularly for changes.