Using ChatGPT, Midjourney, Dall-E 2, Stable Diffusion and other AI applications in the production of advertising campaigns is super convenient and can save a lot of time and money. However, such use is not without legal risk. Not surprisingly, many ad agencies and international advertising networks have put the use of AI on hold, banned it, or subjected it to strict rules. But what legal risks are we really talking about? Without pretending to be exhaustive, in this post I will mention two relevant issues in the field of Dutch copyright law.

1.    It is often unclear whether - and if so, to what extent - the output of AI is protected by copyright

Copyright protection of a work (e.g., a piece of copy) requires human creative input, and that creative input must be evident from the work itself. The output of AI when using little detailed prompts is - under Dutch copyright law - presumably not copyright protected. The output in such a case does not demonstrate human creativity. The more the output is dictated by prompts, the more likely it is that human creative input is apparent in the output, and the greater the chance of copyright protection. Specifically, if ChatGPT is asked to write copy, where a creative text conceived by a human copywriter must be reflected in the output, then of the output at least the text conceived by the human copywriter will be protected. If ChatGPT is prompted to write copy that must fit within a detailed creative format, and the output does indeed express the format in question, then to that extent the output is copyright protected. Similarly, if the output of AI is creatively adapted by a human, the output will likely be copyright protected, seeing that the adapted work will likely show human creative choices. But the boundaries between what is protected and what is not are blurred.

This can be a problem. Clients often expect to obtain the copyrights to the work created by the agency. Indeed, client/agency agreements usually contain a transfer of copyright. Often third-party intellectual property rights are exempt from this transfer (think, for example, the rights to stock materials), but an exception for AI-generated work that forms part of the agency-created work is not (yet) common. Practical tip: if an agency uses AI in the production of advertising materials, it is not a bad plan to dust off the client/agency agreement to see what it says about copyright.

2.    AI's output may unintentionally infringe on others' copyrights

Copyright infringement occurs when a work is derived from another, original work, and that work incorporates so much creativity from the original work that the work and the original work are similar in overall impression. Derivation can be readily assumed in the case of output from AI. The output is based on (and thus by definition derived from) existing work that was used as training data for the AI application. If the output of AI is very similar to an existing work, because of similar protected elements, then the risk of copyright infringement is high. Therefore, the use by agencies of (unedited) AI output in work created for clients is risky. If the client publishes the work in question, and a copyright claim pops up, the agency is liable (except if something else has been agreed with the client).

Conclusion

Copyright law gives every reason for caution in the use of AI output by advertising agencies, especially in advertising materials intended for actual use. It is advisable to be completely open within the agency and in the relationship with the client about the use of AI, and to make proper agreements about the allocation of risks. 

Daniel Haije

PS: This post came about without the use of generative AI :-)

4 October 2023