What does the Bartz v. Anthropic case mean for Serbia: copyright, AI, and new regulation
The Bartz v. Anthropic case took place before a US court, but its consequences are global. For Serbia, which wants to be a regional technology hub and at the same time a country that protects authors and creatives, this is an important signal.
This text deals with where Serbia is today, where it is heading in terms of artificial intelligence regulation, and what domestic companies, authors, and lawyers practically need to do.
Where we are now: Copyright and Related Rights Act
The Serbian Copyright and Related Rights Act is based on the classic principles of continental copyright law:
- The author can only be a natural person;
- A copyrighted work is the author’s original intellectual creation;
- Copyright arises by the very act of creation, without formal registration.
Within that framework:
- Artificial intelligence cannot be an author – it has no legal capacity, nor does it possess “intellectual creativity”:
- The programmer, client, or user of an AI system can potentially be considered the author if their contribution is creative and meets the conditions for authorship.
However, the law does not contain explicit rules on:
- Whether and under what conditions it is permitted to use already protected works (books, articles, photographs, music) for training AI models;
- How mass downloading of content from the internet for the purpose of training models is treated;
- What rights authors have whose works were used in training without their consent.
Because of this, the domestic legal situation on this issue is currently unclear. There is no developed case law on AI training, nor specific provisions in positive law that would directly answer these questions.
Low awareness and lack of practice
Another problem for the Serbian context is the relatively low level of awareness of copyright issues and the limited number of court proceedings in this area, especially in the digital environment.
This means:
- Authors and publishers are often not even aware that their works may already have been used to train foreign AI models;
- Domestic technology companies, in their desire to “follow trends”, sometimes download ready-made datasets without analyzing the legal status of the content;
- The courts still need to build practice and expertise to deal with complex cases where copyright and new technology intersect.
In such a situation, international precedents such as Bartz v. Anthropic have special importance because they offer guidance on what is considered acceptable and what is not in developed legal systems.
Where we are heading: Future artificial intelligence law and alignment with the EU
Serbia has already announced that it is working on a separate artificial intelligence law that should be aligned with European trends, primarily with the EU Artificial Intelligence Act (EU AI Act).
Plans envisage:
- That the draft law will be prepared in the coming period;
- That by the end of this decade, a functional framework for regulating AI systems (including high-risk ones) will be established;
- That the Serbian system will gradually approach EU standards.
The EU AI Act, although primarily focused on the safety, risks, and transparency of AI systems, also contains provisions that directly concern copyright. Particularly prominent are:
- The obligation for model providers to ensure compliance with copyright rules during training;
- The obligation to publicly disclose concise descriptions of the content used to train models;
- The obligation to respect “reservation of rights” mechanisms (e.g. content owners can clearly indicate that they do not allow their content to be used for AI training).
Although Serbia is not formally an EU member, it is clear that any domestic AI company targeting the EU market must count on these standards being practically applied to it.
What the Anthropic case means specifically for Serbia
From Serbia’s perspective, the Bartz v. Anthropic case sends several very concrete messages.
For AI companies in Serbia:
- It is necessary to immediately begin an internal “inventory”: where does the data for training models come from;
- .Using pirated sources (regardless of whether the servers are abroad) is a high legal risk. If the model operates in the global market, the question of jurisdiction will not protect the company;
- It will be necessary to develop a clear policy of respecting copyright, including acquiring licenses, using the public domain and open-licensed content, while complying with the conditions;
- Documentation on the origin of data will become an important part of the company’s legal protection – what today seems like a “technical detail” can tomorrow be key evidence in a dispute.
For authors, publishers, and other rights holders:
- It is time to think about their own strategies: whether they want their content to be licensed for AI training and under what conditions;
- Organizing through associations, collective organizations, and professional bodies can significantly strengthen their negotiating position vis-à-vis large AI companies;
- The value of works in the AI world is no longer only in classic revenues (sales, subscriptions), but also in the possibility of licensing them for model training. The Anthropic settlement shows that this value can be very concrete.
For lawyers and lawmakers:
- It is necessary to monitor global practice and think ahead, rather than waiting for disputes to arise and then reacting;
- When drafting future regulations on artificial intelligence, the issue of copyright must not remain on the margins – on the contrary, it must be one of the pillars;
- It will be necessary to define at least basic principles: when AI model training is allowed, when licenses are required, what rights authors have whose works enter training, how to treat model outputs that are similar to originals, and the like.
Practical advice for the Serbian context
For AI companies:
- Do not wait for the law to “catch up” – create internal policies in line with best international practice now;
- Avoid datasets of unknown origin or those that obviously come from pirated sources;
- Conclude licenses with rights holders wherever possible and economically justified.
For authors and publishers:
- Inform yourselves about how AI systems work and how they use content for training;
- Consider whether you want to enable licensing of your content and under what conditions – a complete ban and full approval are not the only options;
- Consult lawyers about options for protecting your interests in this new sphere.
Conclusion: Serbia at a crossroads
The Bartz v. Anthropic case shows that copyright and artificial intelligence are inextricably linked. For Serbia, this is both an opportunity and a challenge:
- An opportunity to position itself as a country that develops responsible and legally compliant AI;
- A challenge to adapt the existing legislative framework to a reality in which data and content are the key raw material for new technologies.
The sooner we recognize these trends and adapt, the more prepared our authors, publishers, technology companies, and lawyers will be for a world in which artificial intelligence and copyright are in constant but constructive tension.
Need legal advice on copyright and AI?
Follow for more legal insights:
