Intellectual property and agreements with AI integrations: the new terrain of “copyleft” and output rights.
The increasing incorporation of artificial intelligence (AI) models into SaaS solutions and international business environments is reshaping the rules of intellectual property.
In contracts that regulate the use, training or integration of AI systems, the main focus of risk is not only in the algorithms, but in the datasets and the outputs they generate.
The fundamental question is: who owns the results produced by an AI trained with third-party data?
Model and dataset licensing: a critical point in the value chain
In AI integration contracts—for example, a SaaS agreement between a technology provider and a user company—the first step is to accurately identify which elements are licensed and under what terms.
The model may be covered by a proprietary (private) or copyleft (open source) license, while the datasets used in its training may have disparate origins and conditions: public data, databases licensed under Creative Commons, or internal company datasets.
Some model or dataset licenses — such as Apache 2.0 , MIT or Creative Commons Share-Alike (CC-BY-SA) — impose attribution obligations, distribution under the same terms, or limitations on commercial uses.
These conditions can "contaminate" the outputs if contractual mechanisms are not established to delimit them.
Recommended standard clause:
“The Provider guarantees that the datasets and models used in the Training are not subject to licenses that impose copyleft conditions or restrictions incompatible with the commercial exploitation of the Output generated by the Client.”
Output rights: derivative work or autonomous creation?
The main legal debate today revolves around whether the results produced by an AI (texts, images, designs, code) can be considered works protected by copyright , and who would be their owner.
In the European Union, copyright protection requires creative human intervention ; therefore, outputs generated autonomously by an AI model are not protected unless the user has made a significant contribution to the configuration, selection, or editing of the result.
From a contractual perspective, this ambiguity requires that ownership be expressly assigned.
In international SaaS licensing or provision agreements, it is recommended to incorporate clauses for attribution and differentiated use of the "model," "datasets," and "output."
Suggested standard clause:
“The Parties acknowledge that the Outputs generated through the use of AI will be the property of the Client, without prejudice to the rights that correspond to the Provider over the underlying model and the datasets used for its operation.”
Risk matrix in AI integrations
| Element | Legal risk | Size contractual |
| Third-party dataset | Copyleft or unauthorized use | Guarantee of legal origin and declaration of compatible licenses |
| Pre-trained model | Restrictions on use or sublicense | Disclaimer and Limited Access Clause |
| Generated output | Uncertain ownership | Express assignment of intellectual property |
| International commercial use | Incompatible licenses between jurisdictions | Choice of law and territoriality clause |
FAQs Frequently asked questions on AI contracts
It depends on the degree of human intervention. If the result is generated automatically, it may not be copyrightable. However, it can be contractually attributed to the user or customer, which is advisable in B2B relationships.
Only if it is guaranteed that the datasets used do not impose copyleft or "non-commercial" restrictions. Otherwise, use could be considered infringing. It is essential to conduct a prior license audit and document the origin of the data.
The provider could be subject to infringement for unauthorized reproduction or modification. The contract must include representations and warranties regarding the legality of the training, and mechanisms for redress in the event of third-party claims.
A new contractual governance in the era of AI
Technology procurement with artificial intelligence components requires a more granular approach: distinguishing between model, data and results, auditing the origin of each, and planning for the transfer and use of the output under a clear legal framework.
In international SaaS agreements with integrated AI, clarity in the licensing chain is the best defense against intellectual property disputes and the regulatory uncertainty that still persists.
RRYP Global, intellectual property lawyers.