Dr. Andreas Dustmann, Attorney at Law at BOEHMERT & BOEHMERT

ChatGPT & Co. – Artificial Intelligence and Copyright

/in

Few topics are currently dominating the public debate as much as the impact of artificial intelligence on our workplace. This also applies to copyright law and the creative industries that depend on it. The following article provides a brief overview of the most important copyright issues arising from the use of AI. All issues can be traced back to the same question: “Who benefits from technological advances?”

Conflicts of Interest

In recent years, AI systems of all kinds have become exponentially more important. So-called generative AI systems, such as ChatGPT and Google Gemini for text or Midjourney for images, have attracted particular attention. These are systems that are able to autonomously generate media content based on human input (so-called “prompts”). Their products are not only deceptively similar to works of human creativity but can also replace them.

The generation of new content takes place in 4 steps:
1. Collection of training data in a database (“corpus”).
2. Analysis of the training data for patterns and correlations, followed by “fine tuning”.
3. User command (“prompt”).
4. Autonomous generation of new content.

The diversity of interests is already evident here. The rightsholders of the pre-existing works used for training (“training data”) want to be remunerated for their use or to be able to prohibit this form of use. Operators of AI systems, on the other hand, are dependent on the training data as it is not possible to generate content without it. They stress that overly strict regulatory requirements will make a key technology of the future unattractive. AI models would not replace existing works, but only create new content. Ultimately, users of the systems in question want to be able to use the works generated on the basis of their specifications as extensively and exclusively as possible.

When a weeks-long strike by screenwriters and actors paralyzed the Hollywood film industry in the summer of 2023, the filmmakers made one demand in particular: their work should not be replaced too much by the use of artificial intelligence (AI) in the future. In the future, specially trained AI models will not only generate scripts and individual scenes, but entire movies.

Is AI-generated Content Protected by Copyright?

Long before ChatGPT, the question of whether AI-generated content – text, photos, software, music – is protected by copyright, and if so, who is entitled to that protection, was debated. This question can be considered settled: according to the creator principle enshrined in almost all copyright laws, copyright protects only the results of human intellectual creation. Machine-generated works, which are created completely autonomously without human intervention, do not enjoy copyright protection. According to this understanding, the work of an artificial intelligence cannot be protected because it is the machine that creates, not the human. Only if the user gives the AI such specific instructions that the design of the work has already been determined, and the AI only has to implement that design, would copyright protection be justified. These requirements are generally not met by today’s text and image generators. They create content autonomously and are ultimately unpredictable.

It has often been argued against the lack of copyright protection for AI products that users of AI often invest considerable effort and creativity in designing a prompt in such a way that the AI model generates the desired content. For example, in a case before the US Copyright Office, the author of a 16-page comic featuring Zarya of the Dawn, whose images were generated by the image AI Midjourney, argued that it had taken her almost a year to develop appropriate prompts and procedures to obtain the desired image using the image AI. In a very readable decision of 21 February 2023, the US Copyright Office rejected this argument, stating that no amount of preparation could change the fact that the photo was created autonomously by the AI.

However, content created by AI can be protected by copyright if it is subsequently edited and developed by humans. However, this requires that the adaptation itself is a personal intellectual creation and is not limited to editorial and technical adaptations. Strict requirements must be met. Creators who are dependent on the exclusivity of the content they have created will nevertheless seek to subject pure AI works to their own creative adaptation to obtain copyright protection.

Protectability of the Prompt

From the point of view of the user of an AI system, the most important input is the prompt, i.e. the instructions to the AI. Good prompts are the basis for good AI output, and a new profession, the prompt engineer (“prompter” for short), has emerged to formulate these instructions. This raises the question of whether prompts are protected as literary works. The answer is yes, provided that the prompt itself meets the requirement of personal intellectual creation. The content generated by the AI model using the prompt is not taken into account in the assessment.

Does the AI Product Infringe Third Party Rights?

The fact that the AI product itself is not protected by copyright does not mean that its distribution cannot infringe the rights of third parties, e.g. the authors of pre-existing works. An important motive for using AI models is to create new content by referencing or using pre-existing works (examples: “Write me a new story with Pipi Longstocking” / “Create a picture in the style of Keith Haring”). Generative AI can identify typical elements of an artist and imitate the recognized style to create a new image that looks like an artist’s image. Similarly, sequels to novels, characters or films can be created.

Whether this content infringes the rights of the authors concerned will always be a question of individual cases but is based on traditional copyright principles: An artist’s style alone is not copyrightable; only the specific design of a work is copyrightable. The underlying style remains in the public domain, so that the artist does not monopolize his/her chosen cultural technique through copyright, thereby restricting the artistic freedom of others. A different decision will have to be made if the AI product refers to specific works and these are recognizable or essential elements do not fade away. An example of this is the well-known AI work “Vincent Creating the Starry Night on His Laptop” by AI artist David R. Smith. The reference to van Gogh’s Starry Night is obvious in that the work does not fade. If the works of van Gogh, who died in 1890, had not long been in the public domain, the AI image would infringe the rights of van Gogh or his legal successors as an unauthorized adaptation (Section 23 of the German Copyright Act, UrhG).

  Vincent Creating the Starry Night on His Laptop, David R. Smith

If the AI output is a (recognizable) reproduction or adaptation that infringes the copyright of a third party under these standards, the user must obtain a license from the third party for publication or exploitation or be able to rely on a statutory restriction such as pastiche under Section 51a UrhG.

Use of Existing Works as Training Data

Both small and large AI models are based on training with existing works. In the case of large AI models such as ChatGPT, this is assumed to be almost all the content available on the World Wide Web. The AI then independently analyses the entire volume of text for patterns and correlations to determine the rules on which the language is based (“pre-training”).

To put it bluntly: Generative AI only works as well as it has been trained and analyzed with the existing works.

As a result, authors and rights holders are demanding a ban on the use of their works to train AI modules, or appropriate compensation. Numerous lawsuits are already pending in the US by prominent authors, publishers and image database operators who believe their rights have been infringed by well-known AI models. The outcome of these cases remains to be seen. In any case, the treatment of pre-existing works by AI models is the key issue.

In Europe, however, the EU legislator has given the operators of AI models the right of way. With the Digital Services Market (DSM) Directive, a barrier regulation for machine learning will be introduced in 2019, which will be implemented in Germany through Section 44b UrhG: Section 44b UrhG permits the reproduction of works that are necessary for the automated analysis of these works, unless the rights holder has expressly reserved this type of exploitation (“opt out”). However, due to technical limitations, the possibility of opting out currently plays little role. This means that there is a legal authorization to collect works on a large scale and to create a training corpus from them – without the restriction to research institutions, i.e. also for commercial purposes, provided for under the old legal situation in Section 60d UrhG. The aim of this regulation is to promote technical innovation in the EU and to provide companies with a secure legal framework for AI applications.

However, rights holders are opposed to this interpretation. They point out that the application of any limitation rule must meet the requirements of the three-step test laid down in several international copyright treaties, according to which the application of the limitation must neither impair the normal exploitation of the work nor unreasonably prejudice the legitimate interests of the author. If an AI makes it possible to paint pictures in the style of an artist or to create serialized stories, the application of the exception undermines the basis of exploitation and the livelihood of authors.

The outcome of this discussion is open – the pressure on legislators worldwide to create a (new) legal framework for the use of pre-existing works is growing.

Conclusion

Many of the traditional principles of copyright law can easily be applied to the models of generative artificial intelligence. However, they do not solve the core socio-political problem: how to deal with the fact that creative content – text, images, graphics, software and images – which has traditionally been the responsibility of human creativity and whose exclusive copyright protection has formed the basis of an entire industry, can be generated by AI at the touch of a button? Creative people see their livelihoods threatened. And even if you wanted to give AI products some form of protection that could form the basis for exclusive rights and legal exploitation. This would not change the low value of these products and their reduced licensability.

Robert Bernin, Patent Attorney at BOEHMERT & BOEHMERT

Basically plausible: one year after G 2/21

/in

After the decision of the Enlarged Board of Appeal of the European Patent Office (G 2/21 of March 23, 2023) did not seal the end for all experimental data submitted later to prove a technical effect, the question arises one year later how the Boards of Appeal deal with the guidelines established in G 2/21.

In response to the questions posed by the Technical Board of Appeal (in T 116/18 of October 11, 2021), the Enlarged Board of Appeal had answered as follows:

1. Evidence submitted by a patent applicant or proprietor to prove a technical effect relied upon for acknowledgement of inventive step of the claimed subject-matter may not be disregarded solely on the ground that such evidence, on which the effect rests, had not been public before the filing date of the patent in suit and was filed after that date.

2. A patent applicant or proprietor may rely upon a technical effect for inventive step if the skilled person, having the common general knowledge in mind, and based on the application as originally filed, would derive said effect as being encompassed by the technical teaching and embodied by the same originally disclosed invention.

While the first answer clearly states that evidence, usually in the form of experimental data, may not be disregarded because it was filed subsequently, the second answer requires an assessment of whether a technical effect can be inferred from the original application documents.

A good year after the decision of the Enlarged Board of Appeal was announced, there are many decisions of the Board of Appeal relating to G 2/21 and assessing the assertion of a technical effect.

Consideration of subsequently filed evidence

In T 116/18, from which the decision of the Enlarged Board of Appeal arose, it was concluded on the basis of G 2/21 that the subsequently filed experimental data on the synergistic effect of the claimed insecticides must be taken into account.

In T 1515/20 and T 728/21, subsequently filed experimental data were also taken into account, since the technical effect was clearly disclosed in the original documents.

In T 1989/19, the technical effect of an improved shelf life was recognized as encompassed and embodied in the originally disclosed particle size stability of an active substance micronisate. Moreover, it was the view of the Board of Appeal that an admissibly derivable technical effect always implicitly included the aim of improving this effect, since the skilled person always strives for further development and improvement.

In T 2716/19, the fact that a dependent claim was directed to specific alkoxides was seen as an indication that these could be associated with a technical effect. Furthermore, the skilled person could have deduced the improvement in yield from the yield of an experimental example in the original application. The subsequently filed documents proving the improvement in yield were therefore considered.

In T 885/21 and T 1329/21, a technical effect and, based thereon, an inventive step was recognized on the basis of the subsequently filed experimental data, since the technical effect was encompassed by the original documents and embodied by the originally disclosed invention. In these decisions, the technical effect could be directly derived from the original application documents.

In T 1445/21, subsequently filed experimental data were taken into account, even if it was ultimately not sufficient for a technical effect of the distinguishing features. It is noteworthy that the recognized technical effect of olfactory performance or intensity was considered to be encompassed by the originally disclosed technical effect of the stability of the fragrance microcapsules.

Partial consideration of the subsequently submitted evidence

In T 2046/21, subsequently filed experimental data for a technical effect, disclosed in the original application, of a combination of active ingredients were taken into account, whereas further claimed effects according to other subsequently filed documents were not considered, since these specific technical effects were neither disclosed nor indicated in the original application documents. Overall, inventive step was denied.

In T 0681/21, subsequently submitted experimental data showing an originally disclosed technical effect for a preferred feature combination according to the auxiliary request was allowed, while a synergistic effect claimed in the main request for another preferred feature combination was considered not to be covered by the technical teaching of the original application documents. As a result, this synergistic effect could not have been effectively demonstrated on the basis of experimental data submitted later.

No consideration of the subsequently submitted evidence

In T 852/20, the technical effect for the claimed crystalline form of a compound was found not to be covered by the technical teaching, since the relevant paragraphs of the description referred only to the amorphous form of the compound.

In T 258/21, a technical effect based on subsequently filed documents was not considered, since the technical effect was neither disclosed nor suggested in the original documents.

In T 887/21, a technical effect on the basis of subsequently filed documents was not considered, since the technical effect shown there did not even correspond to the vague hypotheses regarding a possible effect in the original application documents. It was emphasized that an invention could not be based on knowledge that only became available after the filing date.

No relevance in case of sufficiency of disclosure

The Boards of Appeal have consistently rejected the transferability to objections in the case of sufficiency of disclosure hoped for by some patent proprietors, particularly in the case of second medical uses (see T 1435/20, T 25/20, T 2790/17, T 1779/21, T 552/22). What matters here is the status on the date of filing and subsequently filed evidence with a later priority date cannot turn the tide on its own. Only in T 1796/22 were subsequently filed documents considered as additional evidence that the application was sufficiently disclosed on the date of filing, in addition to the experimental examples of the original application.

Conclusion

Subsequently filed evidence with later priority may continue to support a technical effect to prove inventive step if this effect can at least be deduced from the original application documents. Applicants are advised not to keep the description of the technical effect of features and feature combinations too brief or even to omit it. Otherwise, this may come back to haunt them later.

Fabio Adinolfi, Rechtsanwalt bei BOEHMERT & BOEHMERT

CJEU: Trade mark infringement by spare parts from third-party suppliers that contain elements for affixing a trade mark emblem

/in

In its judgment of 25.01.2024 (C-334/22), the European Court of Justice (CJEU) clarified that a car manufacturer can prohibit the use of a sign for spare parts if it contains an element that is similar or identical to a trade mark.

Read more
Dr. Lars Eggersdorfer, Rechtsanwalt bei BOEHMERT & BOEHMERT

File Inspection at the UPC

/in

If a company is concerned that a patent may be (or has been) asserted against it, access to the files of parallel validity or infringement proceedings relating to that patent will save time and money in preparing a defense.

The files in opposition proceedings before the European Patent Office or the German Patent and Trademark Office are usually open to the public without restriction, so that the prior art submitted and the arguments on the interpretation of the contested patent and its (lack of) validity can be ascertained immediately. In nullity proceedings before the German Federal Patent Court, access to the file must be requested. Although this is usually granted in full, the patentee can regularly delay the inspection for several months from the filing of the request by means of a mere pro forma objection. In infringement proceedings, on the other hand, German courts regularly do not grant access to the file; only their opinions are published, so that non-infringement arguments have to be awaited.

What is the situation before the Unified Patent Court (UPC)?

According to Rule 262.1(b) of the Rules of Procedure of the UPC (RoP), unlike decisions and orders of the court (Rule 262.1(a) RoP), pleadings and evidence (in all types of proceedings) are not to be made public but must be made public upon a reasoned request and after hearing the parties. Regarding formalities, the Court of Appeal has already ruled (order of February 8, 2024, file no. UPC_CoA_404/2023) that such a request must be made by a lawyer admitted to practice before the UPC.

It was now eagerly awaited which substantive reasons justify the application for file inspection. There had been divergent decisions on this at the first instance level. The Nordic-Baltic Regional Chamber had in principle allowed a generally expressed professional or scientific interest to suffice (order of October 17, 2023, file no. UPC_ CFI_11/2023), while the Munich Central Chamber did not (order of September 20, 2023, file no. UPC_CFI_1/2023 and order of September 21, 2023, file no. UPC_CFI_75/2023). However, both chambers had demonstrated legal doctrinal depth for their respective lines of reasoning and, in particular, derived various arguments from the systematics of the Rules of Procedure and their history. It therefore appeared open as to how the Court of Appeal would position itself.

In its decision of April 10, 2024 (file no. UPC_CoA_404/2023), the Court of Appeal has now basically followed the line of the Nordic-Baltic Regional Chamber, according to which a generally expressed interest in accessing the file is sufficient as long as it is not abusive (paras. 43, 44 and 55 of the decision). The court did not indicate when abuse is given, but since practically any expressed professional or scientific interest is sufficient, abuse constellations should be limited to absolutely exceptional cases. The court justifies this broad approach primarily by stating that business secrets and personal data can be protected by means of a separate procedure in accordance with Rule 262.2 RoP (paragraphs 45 and 46).

However, the fact that the specific case concerned proceedings that had already been concluded could be seen as a limitation of this broad approach. The Court sees no need to preserve the integrity of the proceedings by refusing access to the file where the proceedings have already been concluded whether by a judicial decision or by a settlement (paragraphs 48, 49 and 52). However, the statements in para. 53 then indicates that, in the case of ongoing proceedings, a personal interest in the patent, such as the threat of a claim arising from it, may be necessary to justify a request for access to the file.

In practice, this means that companies may be able to obtain access to the file anonymously through their lawyer in the case of closed proceedings, because a general interest can be invoked, but in the case of ongoing proceedings, the potential personal interest may have to be disclosed – which may be disadvantageous if, for example, there is a fear that the request for access to the file may alert the patent proprietor to a previously unknown patent infringement.

Companies must therefore carefully consider, with the help of legal advice, whether to request access to files in ongoing proceedings and how exactly to justify the request.

Dr. Alexander Thamer, Rechtsanwalt bei BOEHMERT & BOEHMERT

No Design Protection for Rihanna’s Puma sneakers

/in

On March 6, 2024, the General Court of the European Union ruled that a design for a shoe registered by Puma in 2016 was rightly declared invalid by the EUIPO. The design had already been disclosed in 2014 and therefore lacked novelty: On the occasion of her appointment as Puma’s new creative director, Rihanna wore on several photos published on her Instagram account the very sneakers from the planned joint collection that Puma wanted to protect 2 years later by the (now invalid) Community Design.

Rihanna’s Post

On December 16, 2014, the US pop star Rihanna published several photos on her Instagram account to mark the signing of a contract that made her the new creative director of Puma. Also on display on these photos: The white sneakers that Rihanna wore at the signing, with conspicuously thick soles and characteristic ribbing. Perhaps a small teaser for the new collection that Puma would be selling in collaboration with Rihanna. Or a mistake after all? In any case, Puma only registered a Community Design that is strikingly similar to these shoes on July 26, 2016.

A registered Community Design is only protected if it is new and has individual character before the date of the filing of the application. This means in particular that the relevant specialist circles must not have been aware of an identical design at that time. However, the law grants a grace period of one year. If the designer of the design himself or a person associated with the designer discloses the design, this disclosure is not taken into account for applications within the next 12 months in accordance with Art. 7 para. 2 CDR.

However, in the present case this deadline would also have passed for the application of July 26, 2016 if the Instagram photos of December 16, 2014 were indeed a prior disclosure of the design.

A third company, which was involved in a legal dispute with Puma in the Netherlands, promptly filed a request for a declaration of invalidity of the Community Design with the European Union Intellectual Property Office (EUIPO) on July 22, 2019, arguing that the contested Community Design lacked novelty and individual character due to the prior disclosure.

After the EUIPO granted the request and Puma appealed against this decision, the General Court (EGC) now had to deal with the facts of the case.

The Decision of the EGC

Puma objected to the request and the EUIPO’s decision on the grounds that the screenshots of the Instagram photos attached to the application were not sufficient to establish a disclosure of the design that would be prejudicial to novelty. In the photos, Puma argued, Rihanna was in the foreground; the sneakers, on the other hand, were hardly noticed. The photos were too dark and blurred for anyone to be able to clearly recognize the sneakers at all. In addition, at the time of the disclosure, Instagram had not yet offered a feature to zoom into photos in the app, which made it even more difficult to see the shoes. However, according to Puma, it could be assumed that Instagram was used almost exclusively on smartphones, where the photos could only be viewed in a small size. Ultimately, it could not be assumed that all design features of the design were already disclosed by the photos.

None of these arguments convinced the court. In the opinion of the judges, the photos were of sufficient quality to recognize all decisive design features of the design. In some of the photos, the shoes and not Rihanna were clearly in the foreground. Even without the zoom feature, it was also possible to take screenshots of the Instagram photos at the time of the post in order to enlarge them outside of the app. Ultimately, due to Rihanna’s high profile, it must be assumed that both her fans and the relevant professional circles would have an increased interest in the clothing she wears in Instagram photos published in the context of the signing of a contract that makes her the new creative director of a fashion company.

Implications

The court’s decision is hardly surprising, but rather confirms the existing case law regarding prior disclosures that are prejudicial to novelty. However, the example also clearly shows that the one-year grace period can often become a stumbling block. This does not only apply to world-famous pop stars. The publication of photos of a new design on the company website or the corresponding social media channels also quickly constitutes a disclosure within the meaning of Art. 7 CDR. If no designs have been registered at this time, you should therefore always be aware that the clock for filing an application can start ticking from this publication.