ChatGPT and Swiss copyright law: a brief and selected overview

Recently launched ChatGPT (for “Generative Pre-trained Transformer”) is on everyone’s lips. This chatbot, developed by OpenAI, an American artificial intelligence (AI) research laboratory, seemingly takes AI to another level and produces text that is hardly distinguishable from human-written content.

A concrete example of its functioning says it all. When asked to “create a 2-sentence poem about the beauty of nostalgia using the style of Paul Auster”, ChatGPT produced the following in under 5 seconds: “In the quiet of memory, where time stands still / The beauty of nostalgia lingers, a bittersweet thrill”. Fairly convincing, isn’t it?

Interesting questions of copyright law stand among the countless others that ChatGPT raises.

One of them is simple to formulate but quite difficult to answer: who owns the copyright on the generated output? Five main options spring to mind: (i) OpenAI, as the creator and owner of ChatGPT, (ii) the user of ChatGPT, (iii) ChatGPT itself, as a (somehow) digital entity, (iv) the creators of the original source content used to train ChatGPT (hereafter referred to as the “sources”) or (v) some (or all) of the above.

Under Swiss law, two of these options are not viable.

Indeed, OpenAI unilaterally declared that it “will not claim copyright over content generated by the API for you or your end users”[1], which is a valid waiver under Swiss law.

As regards ChatGPT itself, it cannot be the owner of any copyright, for two reasons. The first one is, quite simply, that ChatGPT (like any robot) has no legal personality[2]. Thus, it cannot own anything – and in particular not copyright. The second one is that an “author” of a work protected by copyright can only be a human being – which ChatGPT, a digital entity, is not.

The remaining options available are therefore that either (i) the user (human being) of the chatbot and/or (ii) the creators of the sources can be considered the owner(s) of the copyright on the generated content.

• Do the users of ChatGPT own the copyright over the generated content?

For the chatbot’s users to be considered as the owners of the copyright over said content, two conditions must be fulfilled:  

First, ChatGPT’s output must have an “individual character”, i.e., be original. Interestingly, ChatGPT itself seems to express certain doubts as regards its capacity to produce anything original: “ChatGPT (…) is ultimately still generated by a machine using pre-existing data. This means that it is unlikely that ChatGPT’s output would be considered truly original”[3].

Whilst ChatGPT’s modesty is admirable, it is likely that a large part of ChatGPT’s production is in fact sufficiently original to be eligible for copyright protection. That said, if two different users enter the same input and obtain an identical output from ChatGPT, which is possible[4], only the first output will be deemed “original” – and thus protected.

Second, the AI-generated output must qualify as an “intellectual creation”, namely an expression of a statement of thoughts. As a result, AI-generated works may only be protected by copyright if a human being is sufficiently involved in the creative process[5].

This is where matters get tricky given that the user’s only involvement in ChatGPT’s output is his/her command (“ChatGPT, create a 2-sentence poem […]”).

This therefore raises the question of whether said command is, itself, eligible for copyright protection. Swiss copyright law provides that not only final works of art but also drafts and parts of such works are protected, insofar as they are original enough. For example, not only can the screenplay of a movie be protected by copyright, but so would be much shorter preparatory works such as the exposé and the treatment[6].

Yet, mere ideas are not protectable. For instance, the Tribunal de grande instance de Paris decided in 2005 that the very idea of a short film in which the heroine searches for a person whose photograph is found in a photo booth could not be protected by copyright. Thus, the subsequent movie “Amélie” (also known as “The fabulous destiny of Amélie Poulain”), which depicts a similar story, was not considered as an infringement[7]. The same reasoning would likely apply under Swiss law.

As such, there is no definite answer as to whether a request made to ChatGPT by a human is protected by copyright under Swiss law. Although the more sophisticated a request is, the more likely it will be protected.

Now assuming that, based on a non-original request, ChatGPT produces original output, what happens? Given that the sole involvement of the (human) user in the creative process is not original, ChatGPT’s output cannot be protected by copyright.

• Do the creators of the sources own the copyright over the generated content?

For the creators of the sources to own the copyright, three conditions must be met:

First, ChatGPT’s output must be original enough to benefit from copyright protection. The question is the same as the one addressed above.

Second, the sources must also be original enough to benefit from copyright protection.

Third, the sources must be identifiable in the AI-generated output. If this is the case, the latter qualifies as a derivate work, which requires the consent of the creator(s) of the sources. There is no derivate work when the original characteristic of the sources gives way to the originality of AI‑generated output, i.e., when the sources are no longer, or barely, recognizable[8].

Whether a source remains identifiable in an AI-generated output thus depends on the circumstances of each case. But the more ChatGPT rephrases, alters and mixes its sources of inspiration – in other words, the more ChatGPT is genuinely creative – the less likely this will be the case.

[1] https://help.openai.com/en/articles/5008634-will-openai-claim-copyright-over-what-outputs-i-generate-with-the-api

[2] See the Guidelines on Artificial Intelligence adopted on 25 November 2020 by the Swiss Federal Council p. 8.

[3] See the example reported on the blog of Priori Legal, available at https://www.priorilegal.com/blog/openai-chatgpt-copyright

[4] See ChatGPT’s Terms of use, clause 3b, available at https://openai.com/terms/

[5] S. Ragot et al, “Copyright in artificially generated works”, report of the Swiss Group of the AIPPI, in sic! 2019, pp. 573, 574; Tissot/Kraus/Salvadé, Propriété intellectuelle, 2019, § 16 p. 7.

[6] E. Willi, Le nouveau droit d’auteur, 2021, N39 ad art. 2 LDA; D. Meier, Urheberrechte bei Ko-Autorschaft am Drehbuch, sic! 2018 p. 134, 136; M. Annatina, Interessenausgleich im Filmurheberrecht, 2008, p. 64.

[7] TGI Paris, 11 February 2005, n° 03/08082.

[8] E. Willi, Le nouveau droit d’auteur, 2021, N9 ad art. 3 LDA.