Generative artificial intelligence (GenAI) technologies are transforming industries by enabling machines to create content ranging from text and images to music and videos. As businesses increasingly adopt these technologies, it is crucial to draft contract clauses that address the unique aspects of GenAI. This article explores key issues in GenAI contracts and provides sample provisions to help ensure comprehensive and clear agreements.
- How to define GenAI.
Defining what constitutes GenAI within the contract is the building block on which the remainder of a GenAI contract provision is built. This can include specifying GenAI functionalities, capabilities, and limitations. Depending on the situation, a broad or narrow definition may be appropriate- but if a broad definition (such as just stating “generative artificial intelligence”) is used, be sure you understand the breadth of the definition and the implication of that broad definition in interpreting the contract.
Generally, GenAI is considered to be an artificial intelligence (a computer system or algorithm that has the ability to imitate intelligent human behavior) that is capable of generating new content (such as images or text) in response to a submitted prompt (such as a query) by learning from a large reference database of examples.
But this definition can be both overbroad or too narrow, depending on the specific context of the contract in play, so care should be taken in adopting an appropriate definition for the context. A potential definition that is generally applicable can be:
Generative Artificial Intelligence (“GenAI”), meaning an artificial intelligence (a computer system or algorithm that has the ability to imitate intelligent human behavior) that is capable of generating new content (such as images or text) in response to a submitted prompt (such as a query) by learning from a large reference database of examples. GenAI includes, but is not limited to, the use of ChatGPT, Harvey.AI, Google Bard, Suno, Claude, Bing AI, Sensei, DALL-E, GitHub Copilot, Microsoft Copilot, Generative Design, Synthesia, Alphacode, and any similar GenAI services.
- Notice & permission for GenAI use.
The issue of whether notice, and permission, are needed to use GenAI is an important threshold issue. Often, including a notice and consent rubric in the contract will out whether a vendor or other provider intends to use these types of tools. Some arrangements make a notice and permission rubric impractical, and in others, they can be workable and important provisions (such as when dealing with particularly sensitive data). Sample language could include:
Company shall obtain Client’s prior written consent before utilizing GenAI to perform the Services or including GenAI inputs or outputs within the deliverables provided or utilized in performing the Services.
Another provision that could be included on the issue is:
Company shall disclose its use or inclusion of GenAI to Client.
Or yet another variation to restrict certain uses based on geography or training use could be:
Company shall obtain Client’s prior written consent before utilizing GenAI to perform the Services or including GenAI inputs or outputs within the deliverables provided or utilized in performing the Services where any GenAI utilizes any GenAI data hosted outside the continental United States of America or has any ownership or control interest by any individuals or juridical entities located outside the continental United States of America, or where any of Client’s data may be used to train or otherwise modify the GenAI.
- GenAI intellectual property rights.
Contracts must address the ownership of GenAI technology and the content it generates. This can be even more challenging given recent litigation regarding intellectual property rights associated with the inputs to various GenAI providers, making indemnities and other protections particularly important where there can be a sui generis element of risk in the chain of ownership associated with novel GenAI technologies. It is important to delineate who holds the intellectual property rights and how these rights are managed or transferred, and also to consider the intellectual property rights or related inputs.
Unique to GenAI clauses are concerns about data unrelated to a contract- for example, rights given from Company A to Company B for data inputs and training could inadvertently grant training rights from Company A to B on matters unrelated to the contract unless otherwise specified, if the rights were given in a blanket manner. Sample language addressing these rights could include:
Company shall ensure that it has, and represents and warrants that it will have, the appropriate intellectual property rights associated with any GenAI content associated with the Services, and Company indemnifies and holds Client harmless from any claims, suits, and expenses (including the obligation to advance reasonable attorneys’ fees), associated with any claims against Client associated with the intellectual property rights related to the Company’s use of GenAI.
A version with broader indemnity might include:
Company shall ensure, and represents and warrants, it will have the appropriate intellectual property rights associated with any GenAI content associated with the Services, and Company indemnifies and holds Client harmless from any claims, suits, and expenses (including the obligation to advance reasonable attorneys’ fees), associated with any claims against Client associated with the Company’s use of GenAI.
- Data use, confidentiality, and privacy.
GenAI systems often require vast amounts of data, and are hungry to learn from their use. Contracts should specify what data will be used, how it will be sourced, and the measures in place to protect privacy and comply with data protection laws. They should also address any requirements from various privacy laws, with care taken to ensure that space is made for advancements in the law and geographical scope. For example, GenAI related law in the European Union may or may not be a concern for a contract provision depending on the scope of use. Potential language could include:
Company will not use Client data or inputs related to the Services as training data for any GenAI tool except if such use is essential for the performance of Services, and where Company has given written consent to the use of the GenAI tool; if the Services require the use of Client data as GenAI training data, Company shall ensure that such use is only for the benefit of Client, and that the Company and Client have the ability to control and delete Client data.
Other language ensuring protection from spillover rights could say:
Company further agrees that any permitted Client data and inputs shall be strictly limited to such data and inputs made available under this Agreement, and shall not extend to other Client intellectual property (including but not limited to Client’s film, television, consumer products, home entertainment, and interactive games).
- Liability and indemnification.
Given the potential for errors or misuse of AI-generated content, it is crucial to establish clear liability and indemnification clauses. These clauses should outline each party’s responsibilities and protections in case of legal claims. These provisions will generally be included in standard limitation of liability and indemnity clauses, but for a stand-alone broad GenAI indemnity, language could include:
Company indemnifies and holds Client harmless from any claims, suits, and expenses (including the obligation to advance reasonable attorneys’ fees), associated with any claims against Client associated with the Company’s use of GenAI.
- Performance standards and compliance.
Where the use case is specific, consider setting clear performance standards for the AI’s functionality. This will change radically for each contract, but for a marketing content creator using GenAI, a sample provision could include:
- Termination and force majeure.
Specify the conditions under which the contract can be terminated, including breaches of terms, non-performance, or changes in applicable laws and regulations, is key to GenAI. For example, changes in law could materially change the parameters (or even ability) to use GenAI should be considered, especially whether to consider those types of scenarios a force majeure event (and if so, whether to allow termination in the event of GenAI legislation or case law preventing the essential terms of the contract).
- Ethical use and bias mitigation.
Consider the inclusion of provisions that address the ethical use of GenAI and strategies to mitigate bias in GenAI. This ensures responsible deployment and usage of the technology, and compliance with the same- sample language can include:
Company shall review any auto-generated GenAI content to ensure accuracy and that there is no unlawful bias associated with the content.
- Monitoring and reporting.
Depending on the use case, consider requirements for monitoring the AI’s performance and regular reporting to ensure transparency and accountability.
- Maintenance and updates.
Where applicable, outline responsibilities for maintaining the AI system and providing updates or improvements to ensure its continued functionality and security.
- Service level agreements and uptime.
Depending on the GenAI usage, consider addressing service levels in an integrated or separate appended service level agreement (“SLA”). An SLA establishes clear expectations and benchmarks for service delivery, ensuring both parties understand their responsibilities and the standards to be met. SLAs delineate specific metrics for performance, such as uptime, response times, and issue resolution, which helps in maintaining a consistent and reliable level of service. They provide a framework for accountability, enabling clients to hold service providers to their commitments and seek remedies if standards are not met. By clearly outlining the consequences of failing to meet specified criteria, SLAs also tend to motivate service providers to maintain high performance levels, ultimately contributing to smoother operations and better business outcomes.