Liability for copyright infringement caused by AI

AI (Artificial Intelligence), as it is not a separate entity subject to copyright law, is not liable for the content it generates. This is however not obvious and will probably continue to raise doubts in the future. But even today voices can be heard (such as the EU legislator's proposal for an “electronic entity”) claiming that along with its development AI will acquire legal personality and thus it will be liable for damages in case of any harm.

Likewise, given the unequivocal definition of “creative work” in Polish copyright law, which is clearly understandable and generally does not raise any doubts, works created by (or with the use of) AI will not be classified as creative works. A work subject to copyright is every manifestation of creative activity of an individual nature, expressed in any form, regardless of its value, intended purpose or manner of expression [1]. “Creative activity” is key here. Creative activity is generally understood as an intellectual, generative effort leading to the creation of a specific result. Given how AI works and its limitations and legislation – this definition can be currently fulfilled only by a human being.

All this (the fact that AI is not an entity holding copyrights and that the works it generates are not protected by copyright under Polish copyright law) does not mean that the use of AI and the use of content it generates, such as graphics, music, or texts, may not lead to an infringement of copyright. It will be obvious to state that generative AI may lead to plagiarism or visibly use fragments of existing works or imitate the style of a particular artist.  All in all, AI is widely used for those purposes, for example, when it generates works that imitate famous painters. It should be borne in mind, however, that in most cases where such works are generated, copyright protection applicable to the works of such painters has long expired due to the lapse of the statutory period (the beginning of which is most often counted from the date of the author's death). 

Importantly, the most popular AI systems currently do not verify the generated result in terms of whether it resembles the work or uses fragments or the entirety of a copyrighted work, or whether it infringes upon copyrights. Hence, the risk and burden of verifying whether a generated result infringes upon copyright generally lies with the user of the AI tool.

If copyright is infringed upon, Polish copyright law grants the aggrieved party a range of rights, for example, the right to demand:

Acknowledgment and enforcement of the claims may thus have significant financial implications.

In extreme cases, copyright infringement may also involve criminal repercussions. This is because Polish copyright law provides for a sanction, for example, for the misappropriation of another person's work, providing misleading information about the authorship, unlawful recording or dissemination of the work or the circulation of illegal copies of a work [3].

Both the civil-law claims and the criminal repercussions must be directed at the infringing or wrong-doing person, whereas, under civil law, liability is independent of fault.

This raises the question of who will be liable for copyright infringement in the case of AI: will it be the developer of the AI system, the downstream entity offering the AI system to consumers (vendors, distributors, operators), or perhaps the end user? 

At this point, the answer will be case-by-case depending on the infringement, and each case will have to be looked from the perspective of the applicable general civil law, including copyright law. Despite intensive legislative work, mainly at the European level, specific legal regulations for AI liability have not yet been introduced. Worth mentioning is a concept in which the rules for such liability would be modelled on the liability for hazardous products (the latter is based on risk which means that an entity would be liable for the result itself, i.e. for the damage, whether it is at fault or not).

On the one hand, an AI algorithm may be programmed to always infringe upon copyrights, irrespective of the user’s will and actions, and in such case, liability should be borne by the algorithm’s developer. On the other hand, it may be programmed in a way that enables the user to use it without infringing upon third-party rights, and it is the user’s actions, data entered by the user and commands run by the user that lead to generating a work or achieving a result that infringes upon a copyright. In such case, a copyright may be infringed upon without the user's knowledge or intention. It is because it is difficult to assume that, when formulating a command or after obtaining a result, an average user will be able to fully verify whether the result is similar to a copyrighted work, constitutes its modification or uses it in part or in its entirety.

On the footnote, it should be mentioned that an AI system itself (as a computer program) may also be copyrighted (as a work) and using it (without an appropriate license if required) will infringe upon the copyrights of its author.

In all of the above-mentioned cases where copyright is infringed upon by using AI – irrespective of the represented party (the infringing or the aggrieved party) – it is necessary to properly prepare for a possible dispute. It will be important to:

Professional legal services at the pre-court stage significantly improve the odds and greatly reduce dispute duration. It is obvious that a dispute, especially a court dispute, will involve many legal and factual difficulties, including those relating to obtaining evidence given the novel, complex and technical nature of AI and the fact that there is no case law yet to help interpret the relevant provisions. In many cases, it will be necessary to evidence a case relying on an opinion of an expert having an appropriate background so as to establish, for example, how and to what extent AI and the user's actions influenced the end result.

In addition, the analysis of the entity liable for copyright infringement should include a review of the terms and conditions of the various AI tools. The terms and conditions of different tools distribute liability for using the AI work result differently. For example, the ChatGPT terms and conditions state that it is solely the user that bears liability for the use of the generated content, while the Midjourney terms and conditions state that the extent of the user's rights to the creation depends on the version of the tool purchased by the user (in the case of a paid version, the licence obtained allows the use of the creation in the widely described areas of use). The fact that the user pays a fee to the AI developer may be significant when deciding who is liable for an infringement and who should bear the financial consequences. 

Of course, if the purpose of the user or the AI developer is to modify already existing works using the AI tool, assuming that such action will constitute a separate area of use of the work, it will be necessary to obtain the economic copyright in the work or an appropriate license from the creator of the original work.

If, on the other hand, the purpose is not specified and it is impossible to foresee the result of the use of the AI system, it is worth introducing safeguards of the company’s interests; this includes introducing appropriate provisions regarding the rules of use and liability for the use of AI tools in case of a copyright infringement in the applicable contracts with employees, contractors, and business partners. In addition, if AI tools are already being used in the company, it is advisable to audit the adopted solutions to set rules of procedure as securely and comprehensively as possible within the relevant policies and regulations. 

If you are using AI tools in your company or are planning to implement them - contact us.