Copyright in AI-Authored works


Who is the legal author of machine-generated content?
Does the input provider become the author? Or the AI programmer?
And will AI outstrip humans in legal affairs? As always: it depends!

Austrian copyright law is a part of Austrian law that ensures the protection of intellectual property and the rights of creations in need of protection. It protects all works made by the creative effort of the author or inventor, including works of literature, music, visual arts, photography, film, software, computer games and others. It is the right of the author to decide on the use, distribution and publication of his work and copyright law is therefore essential for the protection of his interests.

In recent years, artificial intelligence (AI) has developed rapidly and is influencing how people live, work, and play. This new medium poses completely new challenges for copyright law, because AI systems can now produce self-created works. This raises the question of how to deal with these works and whether they can be protected.

AI is based on algorithms that enable machines to think, make decisions, and perform tasks like humans. This technology can be used in a variety of areas, including the development of autonomous vehicles, the automation of production processes, the recognition of patterns in large amounts of data, and the development of voice assistants. AI systems can analyse images and videos to determine whether or not a work should be copyrighted. AI algorithms can also be used to detect and remove unwanted content such as copyright infringements.

Copyright is a legal form that protects creativity and intellectual property. Although there is no clear answer for AI-generated works, it is likely that they will be protected by copyright in certain cases. In principle, copyright applies to all types of works that can be considered creative and original. However, there is some ambiguity as to whether AI works are also eligible for protection. Since AI algorithms are based on human knowledge and intelligence, there is some debate as to whether they qualify as personal creations. It can be argued that AI works that produce a creative and original performance are in the protected category. These AI works can, for example, be images, music or texts generated by AI algorithms. However, the question of authorship is difficult to answer. Some experts argue that the author is the programmer who developed the AI‘s algorithms. Other experts argue that the AI itself is the author, since it performs the creative and original work.

You might be surprised to learn that this text was not created by a human, but by an AI (artificial intelligence). AI is able to create content in a very short time that is in no way inferior to that produced by human authors in terms of quality and structure. This poses a major challenge for copyright law, because it is becoming increasingly difficult to distinguish machine-generated from human-generated content. Another problem is that no clear line can be drawn as to who is the legal owner of such a text. Should the AI itself be considered the author? Or the AI’s creator? It is therefore important that regulations are created to govern the rights and obligations of AI content creation. This is an extremely complex challenge, but one we must face if we are to protect the creative freedom of all authors and the integrity of copyright.

Would you have noticed that the text above was written not by a “natural” but an “artificial” intelligence, the GPT-3 model “text-davinci”?

For clarity, the AI generated texts are shown in a different font. The “Generative Pre-trained Transformer” (GPT-3) makes it possible to create content in seconds that is indistinguishable from human-created content. The example above makes it clear that this is plausible for general and superficial texts. But can the AI also write legal briefs or even negotiate? Will its explanations stand up to deeper legal – in our case copyright law – examination? Will artificial intelligence outstrip humans in legal affairs?

In addition to aspects of the legal profession that are obviously difficult for a machine to fulfil (just think of the tact required in family disputes or tactical considerations in litigation), the AI lacks depth in reasoning and legal precision. It is also sometimes simply factually wrong. This will be examined in more detail below using the above text as an example.

Perhaps after re-reading the first three paragraphs, some readers may now be inclined to recognise a certain soullessness compared to the work of human authors. This hindsight bias aside, the AI can hardly be accused of anything content-wise until the third paragraph. The explanations are – probably due to the comparatively primitive inputs – both semantically and syntactically faultless.


It becomes more interesting when the first genuine legal assessment is demanded of the AI. The question of whether works created by artificial intelligence are protected by copyright apparently leads the AI to a manoeuvre that is also popular among its human colleagues: “It depends!” The AI is able to locate the crux of the worthiness of protection in the “creativity and originality” of a work and furthermore thinks it can recognise a difference of opinion as to who should be entitled to protection according to the Copyright Act (Urheberrechtsgesetz, UrhG). This is the first place that the limits of AI become clear.

According to § 1, the UrhG protects idiosyncratic intellectual creations. When the AI refers to “creative and original works” in the fourth paragraph, the algorithm does not really use precise legal terminology, but – with a pinch of salt – it is still possible to recognise a large degree of congruence in content between its statements and the wording of the law. This is in stark contrast to the subsequent statements regarding “some ambiguity” and “some debate” about the classification of AI works as personal creations and the associated authorship of AI works. Even with much interpretative goodwill, these are incorrect. Unanimity dictates that a creative act by a natural person is required for copyright to be applicable. This can also result from the more or less concrete instructions (inputs) to the AI (see infobox).

In our case then, the copyright of the generated text (output) would belong to the author (input giver) on the basis of human input, which arguably constitutes a “peculiar intellectual creation”, not least due to the lack of special demands on the “height of the work”. In agreement with the Supreme Court, the situation would be different if “works are created without the intervention of a creative person”. In such a case, the creative work of a human being is missing and the protection of the UrhG does not apply. However, since an AI cannot (yet) program itself entirely, we can conclude from the above that – in addition to the aforementioned “input provider authorship” – the programmer of the AI is also entitled to the copyright of AI works. This renders co-authorship of the input provider and the programmer conceivable.

Although the AI raises this issue, it does not go any further into the background of this debate (which is well worth having) or into its fundamentally correct “it depends” dogmatism with regard to peculiar intellectual creations. Rather, in the last sentence of the fourth paragraph, it strays into a blatant misunderstanding of the legal situation. Instead of discussing the relevant questions (under what conditions does the input provider/programmer become the author), the AI sees authorship of the AI itself as a counter-opinion to “programmer authorship”. However, this no longer has anything to do even peripherally with legally viable argumentation. De lege lata, authorship can only be attributed to natural persons. Opinions to the contrary are philosophical at best.

Two findings can be derived from what has been discussed so far, which somewhat detract from the initial magic of the AI phenomenon. On the one hand, AI does not seem to be able to make direct use of a legal text, which is surprising given the optimal prerequisites of laws as a source (easy accessibility, structured composition). On the other hand, it does not shy away from making factually incorrect statements, and it would be quite interesting to know where the AI obtained this quite obviously incorrect information.

Finally, in the last paragraph, the AI proves once again how advanced the technology already is. It recognises or at least seems to have recognised the intention behind the input for the last paragraph. Even if the wording could have been more pointed, I think that the inclusion of this element of surprise is successful. And even if the AI once again references its supposed authorship of the text, this touch of irony once again succeeds in underlining what was said in the last paragraph. If, in the future, machines can adapt and use nuances (such as a “plot twist”) that were previously considered thoroughly human, they will assume a predominant role in almost all areas of our lives. Essays for homework or seminar papers or even theses will, with precise instruction, become doable by the AI and thus become absurd.


In some industries, this has long been the norm. While, marketing departments and news agencies have been using artificial intelligence for several years to produce product descriptions and short reports, it is clear, at least from this example, that AI cannot currently replace a lawyer. Apart from the indispensable interpersonal aspects, it still lacks the necessary legal depth and reliability of argumentation when it comes to drafting pleadings.

In conclusion: It is important to continue to observe technological developments cautiously, to classify them with the existing (copyright) legal tools and, if necessary, to wait for legislative tightening. Because AI is definitely right about one thing: “AI is a complex challenge, but it’s one we have to face!”