An AI model called ChatGPT has caught the public's attention. Developed by OpenAI, it offers impressive capabilities for generating human-like responses to natural language input. But despite its potential, legal issues around copyright and privacy cast significant shadows over the future of ChatGPT and similar AI models.

A key concern relates to the legal status of the output generated by ChatGPT. According to OpenAI's terms of use, generated content can generally be used freely, as it is either not copyrighted or the copyright belongs to the user. From a copyright point of view, the content generated by ChatGPT can in principle be used freely because it is either not protected by copyright at all or the copyright lies with the user.  However, this is only true under one premise: The output generated by ChatGPT can only be freely used insofar as it does not in turn take over copyrighted content from other works. If the output of ChatGPT contains content to which a third party holds the copyright, these parts are still protected. 

ChatGPT was trained with a vast amount of web pages, articles, books and the like, which were stored in a highly specialized cloud infrastructure. Then, the texts were broken down into individual sentences, and the sentences in turn were cut off at random points to "feed" the model, which was tasked with predicting the missing next word. AI proponents argue that AI training (input) is not copyright relevant per se because it does not ultimately lead to enjoyment of a work. However, there is no consensus in legal practice. 

First lawsuits by book authors in the USA are already pending. Silverman, Golden and Kadrey claim that their works were used without consent and compensation. OpenAI is expected to defend itself in court on the principle of "fair use," among other grounds, against which the plaintiffs may object that their entire work was copied and has artistic value. In addition, OpenAI may possibly take the position that it has used databases from third parties who should actually have clarified the copyright issues. OpenAI also plays the ball further in its terms and conditions by assigning users all rights to input and output as long as they comply with the law.

The argumentation that the training of an AI system is not relevant under copyright law seems risky, especially if one takes as a basis the very broad circumscription of the exclusive right under Swiss law. This formulation is in line with many foreign legal systems in this aspect. 

According to Art. 2 Copyright Act (URG), copyright protects so-called works, i.e. intellectual creations of literature and art that have an individual character. However, in order to be protectable, they must have originated in the human mind and have a certain originality, i.e. individual character (cf. BGE 74 II 112). 

For AI-generated content, it will in future have to be determined in each case, taking into account the specific circumstances of the individual case, whether there is sufficient human involvement in the creative performance for it to be protected by copyright. Where exactly the boundary lies will have to be developed in practice. 

In Switzerland, as in the EU, the US fair use doctrine described above does not exist. However, there are 2 exceptions:

• The first exception is "personal use" - this means that copyrighted works may also be used without the consent of the rights holder in private, school and internal organizational settings (Art. 19 URG). However, the professional training of AI models will hardly ever fall under this so-called personal use. Commercial use is generally prohibited without consent.

• The second exception is text and data mining, but only in the research area (see Art. 24d URG in Switzerland or Art. 3 and Art. 4 Directive EU/2019/790 (EU).

Similar legal challenges could arise in other countries and limit the services of technology companies to further develop AI models such as ChatGPT. Expanding existing law or introducing new laws that regulate the use of AI could be possible solutions.

The legal assessment of these issues is still in its infancy in Switzerland, while the EU is already taking steps to regulate AI through transparency requirements and disclosure obligations. Addressing the legal aspects of AI models such as ChatGPT is critical to ensure intellectual property protection, privacy, and transparency. It is up to companies, developers, and researchers to comply with applicable laws and regulations, while governments and jurisdictions must provide an appropriate and balanced legal foundation that both enables the advancement of AI technology and protects the interests of rights holders and privacy.