AI and copyright
Anyone who creates texts, images or videos using artificial intelligence must also deal with copyright law.

Artificial intelligence creates marketing texts, advertising images, and video films. But are companies allowed to use these creations as they are? When does copyright apply to AI?

For some, ChatGPT & Co are useful tools – others think of AI in terms of copyright and consider the tools to be cheeky thieves. The new generative AI models are trained with content that graphic designers, photographers, journalists and copywriters actually earn money from. Whether it is legally acceptable for artificial intelligence (AI) to use people’s intellectual property is one of the major controversial issues surrounding the new technology.

AI and copyright: All important questions at a glance

Therefore, users who are currently euphorically playing around with AI systems such as ChatGPT or Google’s AI Gemini must ask themselves how they are actually allowed to use the images, graphics and photos generated there. Who is the author of the works? Can you be sued if the AI ​​result used is deceptively similar to a protected text or image?

From a purely legal perspective, there are often no clear yes-or-no answers to these questions. Much is still in a grey area, partly because legislators around the world have yet to classify the new technology.

Is artificial intelligence the author of a work?

No. Only humans can claim copyright for themselves. The companies behind the AI ​​programs have also not provided any protection for the works of their machines – but at the same time grant the user extensive rights of use and exploitation.

AI creations may be used commercially in accordance with the general terms and conditions (GTC) of the respective provider. Open AI, the company behind the well-known language model ChatGPT, for example, gives ChatGPT users all rights to the work. “This means that you can use the content for any purpose, including commercial purposes such as sale or publication,” as the terms and conditions state. However, Open AI also passes the question of copyright protection on to the user, who is then responsible for the content, “including ensuring that it does not violate applicable law (…)”.

In other words, anyone who publishes AI-generated texts, for example, does not have to worry about getting into trouble with the AI ​​manufacturers. However, if the publication violates the copyright of others, for example, you will have to take responsibility for it yourself.

Can AI models be trained with copyrighted works?

Yes – but with restrictions . AI systems must first learn before they can generate content. To do this, companies use freely available information on the Internet. When training generative AI, companies currently do not ask whether others have rights to the content – they simply use it. Legally, this is compatible with copyright law.

Anyone who has copyright to texts, images or music usually has to be asked in advance before their content can be used. However, there is some evidence to suggest that training AI with protected content falls under a new paragraph of copyright law introduced in 2021 ( Section 44b of the Copyright Act ). This allows use if “information, in particular about patterns, trends and correlations” can be obtained from it. Authors must therefore accept this so-called “text and data mining” if their works are used to train artificial intelligence. However, the law also states that the “reproductions” must be deleted when they are no longer required.

“It is not yet clear whether the use of works by AI models falls under data mining,” says Tom Brägelmann, a lawyer in the Berlin office of the Annerton law firm. “Some say that AI copies everything. Others say that it does not create a copy in the strict sense, but only selects content within a neural network – similar to what the human brain does when it remembers an image or a passage of text.”

In the end, the courts will probably decide on this definition question. “When the legislature issued the rules of Section 44b of the Copyright Act, the effects of the new large AI models were not yet known, so one can also question whether these regulations are actually intended for this purpose – even if the wording fits. A court could therefore also interpret this norm in a restrictive or supplementary way,” says Brägelmann.

Can I protect my own content from AI?

Yes. Copyright holders of texts, music or images have the option of opting out of data mining. To do so, they must formulate a so-called “reservation of use” against the reading out – in “machine-readable” form according to copyright law.

In practice, this works via the root directory of the Internet domain. A file can be stored there that determines whether and how the content of the page can be recorded by so-called web crawlers. These are computer programs that automatically search the Internet and analyze pages. Search engines use them, for example, to index websites.

The owner and editor-in-chief of Search One and webmaster pro Kai Spriestersbach explains how such an opt-out works in a blog post . In practice, however, the “opt-out” could also have disadvantages: Text or images may no longer be found by search engines.

Anyone who uses chatbots like ChatGPT themselves will probably have scruples about entering sensitive company content, because the AI ​​also uses this input to train. You should generally not use personal data from third parties here, nor should you use confidential documents – even if providers like OpenAI now give users the option of stopping the AI ​​from training with their own input.

Can the author of the original work sue for copyright infringement ?

That’s difficult. On the one hand, it would be conceivable to sue the AI ​​companies based in the USA that have been trained with copyrighted material. But that would be complicated. “You can of course try to sue ChatGPT or OpenAI in Germany for copyright infringement. The first problem will be that the company doesn’t even have a branch in Europe,” says Brägelmann. A lawsuit would probably have to be served in the USA. “That’s possible, but it’s expensive and can take time.”

On the other hand, authors could take action against anyone who publishes a work that is confusingly similar to their own copyrighted work. In principle, word-for-word reproductions of copyrighted texts can constitute reproduction within the meaning of copyright law ( Section 16 of the Copyright Act ).

If the result is almost identical or at least very similar to the original work, the author can try to sue for copyright protection. Courts then decide on a case-by-case basis whether this is justified. However, the legal clarification of copyright conflicts is often lengthy. An example: “The electropop group Kraftwerk and hip-hop producer Moses Pelham have been arguing for over 20 years about a sound snippet from the song ‘Metall auf Metall’,” explains Brägelmann.

Can I copyright AI-generated works?

This is also rather difficult. For copyright protection, a person’s personality must be reflected in the creation. In this context, lawyers also speak of the “level of creativity” or “level of design” of a work.

The same applies to the so-called “prompts”, i.e. the voice commands with which the user tells the AI ​​what kind of result it should produce. “There can therefore be no copyright on very simple AI prompts,” explains lawyer Tom Brägelmann. Nor on their results in images and text, because they were not produced by a human but by a computer. “There is therefore no ‘personal intellectual creation’ in the sense of copyright,” explains the lawyer.

Basically, “copyright protects the human product, i.e. the concrete expression – not ideas,” says Brägelmann. For example, a certain photo can be protected by copyright as a “photographic work.” However, the idea of ​​capturing a certain motif or imitating the style of a certain art period is not.

An AI work would only be eligible for protection if humans were heavily involved in its creation, for example through particularly creative and complex prompts. “If I ask ChatGPT to write me a sonnet about mashed potatoes that turns out fantastic, it is still too far-fetched to say: With this little prompt I have already created an intellectual work – even if copyright law basically protects ‘small change’,” says lawyer Brägelmann.

In principle, copyright can apply not only to large works, but also to small creations. The dispute between the group Kraftwerk and hip-hop artist Pelham is also about the use of a rhythm sequence that is only two seconds long. The ongoing dispute also shows how difficult it is to balance copyright and artistic freedom. In theory, an AI prompt could also be protected – if it reaches the necessary level of intellectual creativity and is not just for practical purposes.

The “Zarya of the Dawn” case

A legal dispute from the USA illustrates how difficult it is to claim copyright for an AI-generated work: The author Kris Kashtanova wanted to have his comic “Zarya of the Dawn” copyrighted. But later it emerged that he had created the comic with the AI ​​tool Midjourney.

In this case, the US Copyright Office decided that it was not enough for humans to select their favorites from the AI’s suggestions to claim authorship of the work. It therefore canceled the copyright registration. However, because the plot of the comic came from the author, the US Copyright Office issued a modified registration that states exactly what is protected in the comic – and what is not. The text of the comic and the entire comic as a work were registered as a work within the meaning of the US Copyright Act, but not the individual images generated solely by an AI.

The example shows that even with a detailed prompt, one cannot justify copyright protection for the AI ​​output.

Can I enter copyrighted material into the AI?

This can become a problem. It is true that you can edit a work and use the results with the consent of the author. However, if you, as a user, enter someone else’s copyrighted work into an AI program without asking for permission, you are strictly speaking violating copyright.

The question then is whether the AI ​​result is very similar to the protected original – or whether it has been changed by the editing to such an extent that it is no longer protected.

So, as a private user of a chatbot such as ChatGPT or Gemini, do I have to research whether copyright or trademark law is affected by the bot’s output? “In practice, this is likely to fail due to the lack of options,” argues lawyer Brägelmann. “Basically, the burden of proof that the use is OK currently lies with the person who may be using a copyright without permission. In the long term, however, the person who can do so will probably be obliged to provide a copyright justification or license – i.e. the provider of the AI.”

So can I use AI-generated images and other creations without worry?

Yes and no: Using it without checking it carries risks . The user of the AI ​​does not know what content the artificial intelligence was trained with and whether the rights of others were violated. Therefore, he does not know whether the result he achieves with the AI ​​is very similar to a protected work or even identical to it. In practice, it is therefore advisable to weigh up the risks.

“For pragmatic reasons, use for practical texts, whether internal instructions, event schedules, project management instructions, short emails, marketing studies or a press release, is unlikely to ever be pursued in court, even if it means that someone else’s copyright has been used without authorization. The author will hardly notice,” says lawyer Brägelmann. However, caution is still advised.

The more professional creators, i.e. photographers, authors or musicians, engage with AI programs, the more likely copyright infringements will be discovered. Photographers and artists can already use the ” Have I been trained ” tool to find out whether their images have been used to train AI programs.

In addition to copyright, other protective rights, such as trademark or patent rights, could also be violated by the AI ​​programs. Because these are usually registered and can therefore be easily researched via Google, the risk of being sued for the commercial use of AI works also increases. “There will certainly be lawyers who specifically scan this and build a business model on it to pursue corresponding violations. Large companies will certainly do the same in trademark law,” says Tom Brägelmann.

What are the consequences if I violate the copyright of others with AI works?

If in doubt, it can be expensive. There is always a risk of adopting copyrighted or otherwise protected content when using AI creations. The user would then be liable for any copyright infringement.

Depending on how much a court would estimate the user’s fault, the author could face claims for injunctive relief or even damages (according to Section 97 Paragraphs 1 and 2 of the Copyright Act ). The recall of products may also be requested, for example if packaging infringes the trademark rights of another product. The question of which errors would be attributed to the user and which to the provider of the AI ​​remains open.

Where is case law heading?

The fact that AI models such as ChatGPT are trained on a large scale with copyrighted material has long been a source of controversy. There are many lawsuits that are specifically directed against the use of data for AI training purposes. The problem: the authors do not receive any money when their works are used in AI training, but AI companies sell licenses for further use in commercial projects.

The photo agency Getty is suing the image AI company Stable Diffusion in the UK for alleged copyright infringement. The AI ​​company is said to have used millions of Getty image data that were not explicitly made available for training AI systems.

The New York Times is also suing OpenAI and Microsoft for violating their copyrights. Their AI chatbot ChatGPT was trained with data from several million articles in the newspaper. The newspaper is now demanding that the use of the content be stopped and the data already collected be destroyed.

The Axel Springer publishing house is different: it has entered into an agreement with OpenAI for newspapers such as “Bild”, “Welt”, Politico and Business Insider in December 2023. Now the publisher will receive money if ChatGPT uses its articles to answer user questions.

In Europe, creative professionals are campaigning for legislators to take action and better protect authors against AI models. In Germany, for example, artists and authors have joined forces in the “Copyright Initiative”. Among other things, they are demanding appropriate remuneration for creatives from AI companies, as well as more transparency in training data. Some of their demands have already been incorporated into the European AI Regulation ( AI Act ).

What does the European Union’s AI Act bring?

On March 13, 2024, the European Parliament passed the AI ​​Act. It requires large AI base models such as ChatGPT and Gemini to disclose which (copyrighted) material they used to train their models. They must also develop a corporate strategy for compliance with European copyright law.

The new rules will come into force in 2026. However, lawyer Brägelmann considers them to be very lax with regard to copyright issues: protected content is sometimes only addressed in the non-legally binding recitals in the AI ​​Act. “In this respect, the AI ​​Act will only really become relevant in practice once the European Court of Justice has made some clarifications in its first rulings – so in about 5 to 8 years,” he says.

Another unresolved issue is the fact that many activities relevant to copyright, such as AI training and web scraping, do not take place within the EU. “Many copyright regulations in the EU or Germany are therefore not relevant at all,” says Brägelmann.

One of the important questions that courts in the US have to decide is: Does “fair use” count as copyright permission for training an AI or not? So far, this has not been clarified.

From Brägelmann’s point of view, it would make more sense to set up a kind of global rights management company for AI, similar to GEMA or the American Society of Composers, Authors and Publishers (ASCAP), which would then exploit the artists’ rights. But that is still a long way off – until then, many questions about AI and copyright law remain unanswered.

The expert

Tom Brägelmann
Tom Brägelmann is a lawyer at the Berlin law firm Annerton. He focuses on new technological developments and their impact on commercial law. Brägelmann is admitted to the bar in both Germany and the USA.

LEAVE A REPLY

Please enter your comment!
Please enter your name here