An article published by the European Innovation Council and SMEs Executive Agency is trying to clarify the legal implications surrounding AI-generated content, after the voting of the AI Act, highlighting the intersection of technology and intellectual property rights.
By Creatives Unite Newsroom"Artificial intelligence and copyright: use of generative AI tools to develop new content" an article issued by the European Innovation Council and SMEs Executive Agency is addressing the evolving landscape of copyright law with generative AI technologies, which are increasingly being adopted in various creative processes. The text outlines the current legislative framework in the EU, including the implications of the DSM Directive on data mining and the complexities surrounding the ownership of AI-generated works. However, the rapid advancement of AI technology has outpaced current legal frameworks, creating uncertainty about the protection of AI-generated works.
The article identifies potential risks associated with using generative AI tools, particularly regarding the ownership of created content and the commercial exploitation of such works. It underscores the importance of understanding the terms and conditions of AI service providers, which can dictate copyright ownership and rights to exploit the generated content. Lastly, it offers a compliance checklist to help users evaluate generative AI service providers, ensuring that the content created adheres to legal standards and minimizes the risk of copyright infringement.
Artificial Intelligence (AI) is a disruptive technology that has experienced a high level of development in recent years. This explosive growth has enabled its introduction into the consumer market as a commercial product by several software providers. Within this context, AI has become a complete game changer with a high level of sophistication that is expected to make an impact at all levels of people’s lives, particularly in the context of the creation of new original works, which will be our focus here.
In a nutshell, AI is a solution that can be implemented as a tool for many uses, including the generation of new and “original” content ready to be used, modified, or recycled by designers, communication, and marketing teams. Generative AI is capable of producing new types of “original” creations such as texts, images, and videos following prompts introduced manually by the user (the instructions or queries that the user gives to the AI model to obtain a desired result). Those creations are not spontaneous, behind the scenes there is software that implements an algorithm, that had previously been trained with massive amounts of data. Through this training, the generative AI can generate predictions and an association between the input text (the prompts) from the users, and all the pre-existing knowledge learned. With this association, the AI provides an output/ “original” creation, according to the instructions given. The better the AI tool, the better the output fits with the requirements included in the instructions provided by the user.
Employees and corporations have already started to consider the implementation of such AI tools within their business processes, especially for those oriented toward the creative departments and professions where the use of AI as a content creator can play a relevant role. The main reasons are that AI could potentially become a more efficient and economical tool, for either accurately creating new works by investing less resources or expanding and improving the creative style and capabilities of human artists, designers or writers.
Given the speed at which generative AI technologies are advancing, there is still some uncertainty or lack of clarity regarding how these creations are regulated, since law often lags behind technological development, and not many dispositions have addressed the issues related to AI-generated content.
The purpose of this publication is to help you understand the actual legal implications of the generated AI images and to provide a suggested checklist of aspects to consider when evaluating the use of AI software to generate new content, considering the actual legal background and state of the industry.
Two specific issues arise from the interaction between copyright and generative AI, which can be divided into two main categories:
The potential copyright infringement by the developers of these generative AI tools, through the use of copyright-protected materials to train their algorithms. This data is collected through techniques known as "mining".
On the other hand, whether works produced by – or with – generative AI tools are copyright-protected, and who owns the resulting copyright?
The EU institutions have already taken action to address the first aspect. Through the Directive 2019/790/EU on copyright in the Digital Single Market (often known as the DSM Directive), the legal definition of data mining was introduced along with several exceptions or limitations, which only apply when the data/copyright works are used for the training of the AI systems. Specifically, data mining is considered lawful in the EU:
--As an exception for the rights conferred in articles 5(a) and 7(1) of Directive 96/9/EC, Article 2 of Directive 2001/29/EC, and Article 15(1) of the DSM, in the context of reproductions and extractions done by research organisations and cultural heritage institutions for scientific research, assuming they have lawful access to the works and other subject matters intended to be data mined (article 3 of the DSM Directive),
--As an exception to the rights conferred in articles 5(a) and 7(1) of Directive 96/9/EC, Article 2 of Directive 2001/29/EC, Article 4(1)(a) and (b) of Directive 2009/24/EC and Article 15(1) of the DSM directive itself, for reproductions and extractions of lawfully accessible works, when done by any institution, as long as the text which was “mined” was accessed lawfully and the copyright owner has not expressly prohibited the use of the text/work for data mining (article 4 of the DSM Directive).
Both exceptions are subject to article 5(5) of Directive 2001/29/EC, which states that those limitations or exceptions will be only applied for those “certain special cases which do not conflict with a normal exploitation of the work or other subject-matter and do not unreasonably prejudice the legitimate interests of the rightsholder”.
The second aspect – copyright protection and ownership of works created by or with the help of generative AI tools is a bit more complex. Generally speaking, to be afforded copyright protection in Europe (and throughout the world), works and creations must be original (that is to say, they must be the result of the author’s own intellectual work and not a copy of a pre-existing work). Regarding ownership, works are normally first owned by their author with some limited exceptions (such as works created by employees in the normal course of their employment, which are typically owned by the employer). Note that while to some extent harmonised at EU level, copyright law remains to a large extent a matter of national law, and specifics might differ between countries. The above is a general approximation that in a real case may differ from one country to another (Spanish copyright law may vary from the Italian one, and therefore, the ownership regime may be different).
This leaves the question of who owns copyright over a work created by an AI-generative tool with the inputs from a human.
Here we must mention the “Artificial Intelligence Act” (link here), adopted by the EU parliament which will become fully applicable by 2026 and introduces a comprehensive framework to be applied to the development and use of AI tools. While the main text of the law does not touch specifically upon copyright issues, it does repeatedly make specific mention to the topic in the text’s recitals (from 105 to 109) including:
--Recital 105 of the AI Act which reminds of the relevance of the data mining exceptions to copyright introduced by the Directive (EU) 2019/790.
--Recital 107 which stipulates that providers of AI generative models will be required to provide a detailed summary of the content used for the training, in a comprehensive way that will allow copyright or parties with legitimate interests to exercise and enforce their rights under EU law.
Therefore, the AI Act does not tackle the specific issue of ownership over works created through generative AI tools.
The risks that come with using generative AI tools to develop new content
From all of the above, when considering the use of generative AI as part of your company’s creative workflow to create new works (such as images and new designs), two main questions might come to you, which are:
--Ownership of AI-generated content: Who is the owner of all the generated works? The software provider? The company that licenses the software? The employees that use the software?
--Commercial exploitation of AI-generated content: Will my company be entitled to commercially exploit the AI-generated content and stop third parties from using it? Could I implement or combine the AI-generated content into my own creations? Are there potential risks of third-party copyright infringement?
Ownership of AI-generated content: Who is the owner of all the generated works? The software provider? The company that licenses the software? The employees that use the software?
AI generative software cannot create “original” images from zero, it needs a pre-existing background/data input to feed the algorithms through a machine learning model. The origin of those images/input is crucial not only regarding the ownership of the copyrights generated but also to avoid potential right infringement of pre-existing works. In the case that the algorithm is trained with protected works owned by a third party (and protected via copyright, image rights or data protection) without authorisation or without licensing its content, the results may infringe on the rights of these third parties.
Regarding the ownership of the creations made with the use of generative AI tools, the answer will likely vary depending on the following aspects:
--the laws of the relevant jurisdiction that govern the AI and the creation of the work, if any;
--the extent of the role performed by both the human user and the AI platform in generating the output, as having an AI tool fully develop a new work is different from using such a tool to review a work or make slight adaptations to it; and
--the IP provisions under the Terms and conditions (from now T&C) of the license with the service provider.
As a result of those variations, copyright over works developed with generative AI may belong to:
--To the creators of the algorithm who retain ownership over the works created by their algorithm
--To the user of the AI tool (which is the most common)
--To no one (the works created through the generative AI tool are either considered to not be protected by copyright or must be put in the public domain as per the terms and conditions of the AI tool used)
The current standard situation specified at the T&C of the average AI service provider is that the user is considered the right holder of the works obtained. Normally, when ownership goes to the user, the commercial exploitation of this work is also permitted.
However, it may also happen that the software provider wishes to use the created works to feed the AI, to keep the ownership of the works created by the algorithm or just to limit its commercial exploitation. To determine whether this is the case, you should always check the terms and conditions that apply to the use of a specific generative AI tool.
Will my company be entitled to commercially exploit the AI-generated content and stop third parties from using it? Could I implement or combine the AI-generated content into my creations? Do there exist potential risks of third-party infringement?
On the other hand, one of the main concerns regarding the implementation of the AI tools (directly linked to the copyright ownership of the results), is the entitlement to exploit commercially the outputs obtained.
On principle, you are allowed to such exploitation of the results if the user is the owner of the outputs as per the applicable national law and the terms and conditions of the tool used.
However, there are several circumstances where the commercial exploitation may lead to copyright infringement:
--The results obtained reproduce (totally or partially) pre-existing copyrighted works that were used to feed and train the algorithm.
--The T&C specifies that you may not exploit commercially the results obtained.
In conclusion, all the mentioned topics will be governed by the terms and conditions of the specific software provider and indirectly by national/regional laws that shape its content. Therefore, it is recommended to contract with a service provider which ensures that the data used in the algorithm has been trained with legal and licensed content. Additionally, the producer must ensure the free commercial exploitation of the results obtained, and ideally that the results will not be reused to train the algorithm.
