Generative AI: Important Considerations When Navigating Contract Terms

Generative AI offers vast potential for companies to accelerate development and content creation, synthesize knowledge, and eliminate rote tasks, among other opportunities. This article addresses key contractual issues that you may need to navigate when reviewing or negotiating agreements with providers of generative AI tools, with a focus on protecting your company’s business interests, confidential information, and intellectual property under US law.

For purposes of this article, when we refer to “generative AI,” we mean artificial intelligence technology that has been trained to identify relationships and patterns in the data on which the technology was trained (“training data”), and that applies that intelligence to information submitted by end users of the technology (“prompts”) to generate new content such as text, software code, data, images, video and audio for the end user (“outputs”).

There’s a multitude of potential applications for and benefits of using generative AI tools, but there are also potential risks. When considering such benefits and risks, we encourage you to read our Top Ten Considerations for Companies Using Generative AI Internally.

While this article provides information on navigating generative AI terms of use, it is not a replacement for the legal advice of a lawyer. Nor is it a comprehensive review of all relevant issues concerning these agreements – it is meant as a general issue-spotting guide.

Generative AI agreements

Your use of a generative AI tool will generally be subject to an agreement with the provider of the tool (and if it is not, you should consider avoiding use of the tool altogether). These agreements typically take the form of an online terms of service or terms of use, or a separate agreement that you negotiate with the provider, based on their form. They determine the contractual rights and obligations of each of the parties, including who owns and has what rights to use the prompts and outputs, whether those prompts and outputs are subject to confidentiality obligations, and other potentially important legal terms such as indemnities and limitations of liability. In this article, we may refer to such agreements generally as “terms of use.”

Some generative AI tool providers offer an “enterprise” or “business” version with more customer-favorable terms of use than the standard online agreement for general users of the tool. In addition, you may have the ability to negotiate a separate agreement with the provider that more closely aligns to the requirements of your company. In each case, your use may be subject to additional fees, but you should consider whether those fees, when weighed against the risks associated with the generally available tool subject to the “take it or leave it” terms of the provider, are worth paying in order to obtain contractual protections against some of the risks that come with using generative AI.

Furthermore, you should be aware that certain providers of generative AI tools offer different categories of services, such as plugins or APIs that are separate from the “base” version of the tool. These solutions may enable your company to integrate your technology with the tool (e.g., sending prompts and receiving outputs via the plugin or API for an internal or customer-facing integrated service), but also may be subject to different terms of use than the base versions.

Rights in Prompts

Terms of use between the provider of the tool and the end user often state that the end user owns the prompt. However, the terms of use still may give the provider broad rights to use, prepare derivative works based on, and publish or distribute to other users the prompt. In some instances, unless the end user has signed up and paid for an enterprise or business account or otherwise opted out in the applicable account setting, the provider has a contractual right to use the prompt to improve the generative AI tool, including using the prompt as training data. Your company’s employees and contractors must understand whether submitting confidential and/or valuable proprietary data in prompts may indirectly benefit other users of the tool – including your competitors – and/or expose that data to other users. Obligating your employees and contractors to comply with a thoughtful generative AI use policy could help set guardrails around how your company’s information is used in prompts.

Rights in Output

Similar to prompts, terms of use often state that, between the provider of the tool and the end user, the end user owns the outputs, or may otherwise disclaim the provider’s claim to ownership of outputs. However, even in these cases, the terms of use may grant the provider important rights, including rights to distribute or publish the outputs to other users or use the outputs as training data for its AI models. If it is problematic to your company for a provider to possess rights to use the outputs that you generate – for example, because you anticipate that certain outputs may be valuable – you should examine whether the provider offers private enterprise or business accounts or opt-out mechanisms with respect to outputs, or reconsider your use cases.

Confidentiality

Many terms of use do not subject the provider to any confidentiality obligations with respect to the end user’s prompts or outputs. If the version of the tool you are using and the associated terms of use do not contain provider confidentiality obligations, your company will have no recourse against the provider for breach of contract if this information is disclosed. Further, providing third-party confidential information received by you or your company under a nondisclosure obligation may put you in breach of that obligation, and any inclusion of an individual’s personal data in prompts may put you in violation of privacy laws or the contracts or privacy policies pursuant to which you obtained such personal data. Additionally, in the event of disputes between you and other parties regarding your company’s trade secret information, such as appropriation of trade secrets by a competitor, the inclusion of such information in prompts or outputs may be used as evidence that you did not use adequate measures, as required by US trade secret law, to protect the information. If these issues are concerns for your company, you may want to consider a private enterprise or business account with different terms, or otherwise modulate your use of the tool to avoid disclosing sensitive information. Implementing a thorough generative AI use policy should at least put your employees and contractors on notice regarding the risks associated with using confidential information in generative AI prompts or training data.

Indemnities

Many providers’ terms of use do not indemnify end users for any claims, but some providers, especially for enterprise or business accounts, will indemnify for claims arising from infringement of third-party intellectual property (typically excluding the prompts submitted by the end user). On the other hand, many terms of use for popular generative AI tools contain broad indemnity obligations on end users, including for claims arising from their use of the tool, applications and other offerings they develop in connection with their use of the tool, their violation of the terms of use or the provider’s policies, violation of third-party rights by the prompts they provide, and their violation of any laws in connection with their use of the tool. As such, your company’s use of generative AI may result in broad indemnification obligations for the company – but limited or no indemnification recourse against the provider.

Exclusions and limitations of liability

While terms do vary, the scope of exclusions and limitations of liability for many generative AI tools result in little if any contractual liability for providers under terms of use. As is common with click-through agreements, many providers disclaim all liability for indirect, incidental and consequential damages, and some providers disclaim or significantly cap liability for direct damages. Again, providers that offer an enterprise or business account option may agree to a higher liability cap (for example, the fees paid by the end user to the provider). Conversely, end users typically do not receive significant exclusions and limitations of liability under providers’ terms of use. Therefore, unless you have negotiated a more customer-favorable approach with the provider, your and your colleagues’ use of the tool may subject your company to broad liability and, should your use of the tool result in liability to the company, the terms of use are unlikely to offer much protection from the provider.

The use of generative AI tools may present exciting opportunities for your company, but understanding the associated legal risks, along with the contractual terms that will govern such usage, are critical steps in your evaluation of whether to permit use of generative AI, how such tools may be used and which tools to allow. Therefore, we encourage you to conduct a thorough examination of not only the business and technical benefits and risks associated with the generative AI tools your company is considering, but also the legal risks and whether the providers offer more customer-favorable versions of the tools and/or the contractual terms of use that govern those tools.

Many thanks to Jimmy Gilligan, Tom Connors, and Tracy Rubin for their contributions to this article.