Who owns content created by AI? Remember at least these tips!
Generative AI is here to stay. It is already an everyday tool in research, development, and expert work: for example, in drafting texts, data analysis, image generation, and writing code. At the same time, the rapid adoption of AI has raised challenging copyright questions: who owns the outputs created by AI? What kinds of copyright-related risks are associated with using AI?
These themes were explored in an InnoWebinar, which examined the use of AI especially from the perspective of EU legislation and research work. The webinar was hosted by Berggren experts Suvi Julin and Joel Ettanen.
AI regulations are evolving in Finland and globally
Copyright protects the results of creative work, such as text, images, music, software code, and other creative works. Protection arises automatically when a work exceeds the so-called threshold of originality, meaning it is a sufficiently independent and original expression of the author’s creative work.
In research and expert work, copyright-protected material is constantly handled: articles, images, data, and source code. At the same time, regulations in the EU are tightening rapidly. The AI Act, the DSM Directive, and data legislation set new requirements for the development and use of AI systems. Although the AI Act is not a copyright law, it creates a framework for how developers and service providers of AI models must ensure the realization of copyright. At the core are the legality of training data, transparency, and the obligation to address potential infringements.
In Europe, limitations and exceptions to copyright are defined relatively precisely in law. Particularly important is the text and data mining exception, which allows the use of copyright-protected materials for research purposes. This exception may also cover the training of AI, but it does not entitle AI to reproduce protected works or parts of them as such.
Major providers of generative AI are mostly based in the United States, where the approach is somewhat different. A key concept is the fair use doctrine, which allows limited use of copyright-protected works without permission, for example when the use does not commercially compete with the original work. In the context of generative AI, however, fair use assessments are uncertain, and numerous court cases are currently pending.
Who holds the copyright to content produced by AI?
Copyright is fundamentally human-centric. A prerequisite for protection is that the work is the result of a natural person’s creative effort and exceeds the threshold of originality. A mere idea is not sufficient; what matters is the original execution.
Thus, as a general rule of thumb, content produced independently by AI alone does not usually receive copyright protection. Protection may, however, arise for those parts where a human’s creative contribution is clearly identifiable: for example, in the writing of text, structure, selection of images, or final composition. A single prompt is therefore unlikely to generate copyright, but significant human guidance, choices, and post-editing can lead to a protected outcome.
International case law, particularly in the United States, provides illustrative examples. Artist Kris Kashtanova initially received copyright for a comic book. When the copyright office learned that the images had been created using the Midjourney AI, the protection was revoked. In the final decision, copyright protection was granted only for those parts where the human creative contribution was clearly identifiable, such as the textual content and the arrangement of the images.
Can the use of AI infringe someone else’s copyright?
In short: yes, it can – often without noticing. Legal risks may fall on the AI service provider, the user, or even the end client. Infringement does not require intent, and a lack of knowledge does not exempt one from liability.
Copyright infringement may arise, for example, if AI produces content that is too similar to a specific work or parts of it. An example is the case in which, at the end of 2023, The New York Times sued OpenAI and Microsoft. The NY Times claimed that AI language models had been trained on copyright-protected content without permission and that the companies had benefited from this without paying license fees. OpenAI defended itself by arguing that the case involved a rare error in which the examples were obtained through deliberately targeted prompting. The final outcome of the case is still pending, but it has already become a key example of the collision between AI and copyright.
How can risks be avoided?
AI can and should be used, but responsibly, with awareness of risks and with respect for others’ intellectual property rights. Remember at least the following:
Do not assume that you automatically obtain copyright to AI-generated output.
Favor closed AI solutions provided by your organization. Remember that third-party AI models have likely been trained on data that includes copyright-protected (and other intellectual property–protected) material.
Avoid inputting copyright-protected materials into open services.
Critically review all content generated by AI.
Document the role of AI in your work and research processes.
Label generative content as AI-generated when necessary.
Use AI preferably as a support tool (e.g., summarization, formatting) rather than as the primary creator of content.
In short
Generative AI challenges traditional copyright thinking but does not override it. Human creative contribution remains central, and responsibility stays with the user. In research and expert work, AI offers significant opportunities, provided its use is guided by legal understanding, diligence, and good documentation. Time will tell how copyright rules will evolve as AI and its use continue to develop.
Do you want to learn more about the topic? Watch the webinar recording here, and keep an eye on upcoming trainings on InnoBlog!