Artificial Intelligence and Copyright in Artworks
Artificial Intelligence and Copyright in Artworks
Artificial intelligence (AI) currently permeates all areas of life (1). From a technical perspective, AI is regularly defined as the ability of a machine to imitate human competencies such as logical reasoning, learning, or planning (2). Historically, this concept is essentially based on the endeavor to technologically reconstruct human cognitive processes and learning procedures (3). Under European Union law, this technological development finds its regulatory equivalent in Regulation (EU) 2024/1689 ("AI Act"). The latter defines AI systems as machine-based systems that can derive outputs such as content, recommendations, or decisions from received inputs (Art 3 No 1 AI Act). For artists and users of AI systems, this context sometimes gives rise to complex (copyright) legal issues, which concern, in particular, the creator principle (Schöpferprinzip) as well as the substantive protectability of AI-generated "results" (4).
Applicability of the AI Act in Liechtenstein
Due to the obligation under international law to adopt EEA-relevant legal acts pursuant to Art 102 of the EEA Agreement, the AI Act is being integrated into Liechtenstein law, with the Office for Digital Information ("SFID") providing accompanying support to this process (5). While applicability in the European Union is staggered between 2 February 2025 and 2 August 2027, the adoption and the determination of specific applicability for Liechtenstein are currently in progress. Irrespective of this, Liechtenstein providers are already subject to the market location principle (Marktortprinzip) if the outputs of their AI systems are used within the EU territory (6).
The Union Law Framework: Regulation (EU) 2024/1689 (AI Act)
The AI Act established the world's first comprehensive binding legal framework for human-centric and trustworthy AI. The regulation pursues a risk-based approach that establishes obligations in proportion to the risks posed to health, safety, and fundamental rights. Furthermore, it contains elements of product safety law, including requirements for conformity assessments and CE markings for high-risk systems. The addressees are both public and private actors, including providers and deployers in third countries, provided that the system outputs are used in the Union (Art 2 Para 1 lit c AI Act).
Copyright Law and the Creator Principle
Liechtenstein copyright law ties the concept of a work to cumulative prerequisites. Pursuant to Art 2 Para 1 of the Copyright Act (URG), works are exclusively defined as intellectual creations of literature and art that possess an individual character. Such an intellectual creation requires that the work is based on a human will and is the expression of a thought process; it must therefore necessarily be created through human agency (7). Accordingly, under Liechtenstein law, the author is exclusively the natural person who has created the work (Art 6 Para 1 URG). Lacking legal personality as well as human creative power, AI systems are thus excluded de lege lata as holders of original copyrights (8) (9).
U.S. Copyright Office, Cancellation Decision re: Zarya of the Dawn of 21 February 2023
A comparative legal examination of the decision-making practice of the U.S. registration authority demonstrates that, under US law (as well), protectability is mandatorily linked to human creation (10). In the proceedings concerning the comic "Zarya of the Dawn" by Kris Kashtanova, the U.S. Copyright Office held that the images generated by means of the AI technology "Midjourney" did not constitute the product of human authorship ("not the product of human authorship"). Since text commands ("prompts") are to be classified merely as suggestions ("prompts function closer to suggestions than orders") and the user exercises insufficient control over the specific output due to a lack of predictability, the user cannot be qualified as the shaping creator ("master mind"). Protection was recognized exclusively for the text as well as the selection, coordination, and arrangement ("selection, coordination, and arrangement") of the elements as the author's own creative contribution (10) (11).
Copyright Qualification in the Liechtenstein Context
The legal assessment of such factual circumstances is primarily oriented towards Art 2 Para 1 URG in Liechtenstein. Copyright protection of AI-generated individual images consequently does not apply if the output is not directly based on a human expression of thought (7). However, protectability may exist under Art 4 Para 1 URG for collections, provided the selection and arrangement of the elements constitute an intellectual creation with an individual character (12). Internationally, the maxim "no human, no copyright" (currently) remains the prevailing standard, although the requirements regarding the degree of required human contribution vary (13) (14).
Individuality and AI Output
Protection of the AI output thus presupposes that it exhibits an individual character and significantly distinguishes itself from the public domain. This individuality must always be assessed in light of the available scope for design (vgl BGE 130 III 168 Erw 4.3). If a person uses AI merely as a tool and provides their own creative contribution in the process, protection of the result remains possible (16). An intellectual creation, however, only exists if the result remains determined by human will (17). The mere use of predefined prompts without significant controlling influence by the user, on the other hand, does not fulfill these substantive requirements (18).
Copyright Protection of AI Input ("Training Data")
The use of protected works for the training of models fundamentally requires the consent of the rightsholders (Art 10 Para 2 lit a URG) (7). An exemption through the limitation for text and data mining for scientific purposes (Art 24d URG) is disputed in the case of commercially oriented models. This is because it remains questionable whether pattern analysis constitutes a relevant enjoyment of the work (19). To protect their inventories, however, rightsholders may declare a machine-readable reservation of use ("opt-out") pursuant to Art 4 Para 3 of Directive (EU) 2019/790 (20).
Regional Court of München I (LG München I), Final Judgment of 11 November 2025 – 42 O 14139/24 (GEMA/Open AI)
In its final judgment of 11 November 2025 – 42 O 14139/24 (GEMA/Open AI), the Regional Court of München I addressed the impermissible reproduction through the memorization of works in and by AI language models. With regard to the technical dimension of data processing, the Regional Court of München I held that the "memorization" of linguistic works in an AI model constitutes a reproduction within the meaning of copyright law, because the work is thereby "physically fixed" and can "be made indirectly perceptible" (15). The court clarified that the assumption of an impermissible reproduction does not require the identification of a "specifically definable data set" within the model; rather, the decomposition of the work into parameters suffices, provided the "technical retrievability of the original via simply structured prompts" remains intact (15). The chamber thereby strictly differentiates between the creation of the "training corpus" and the "memorization" within the model, whereby the latter exceeds the limitation of text and data mining, as it does not merely serve the preparation of data analysis (15). Liability for copyright infringements through outputs falls upon the operators of the language model as perpetrators, as they exercise dominion over the act (Tatherrschaft) regarding the generation process, as long as the reproduction is not provoked by the user (15).
Transparency Obligations and Enforcement
Based on Art 50 Para 2 in conjunction with Art 113 Para 2 lit a of the AI Act, a comprehensive labeling obligation for synthetically generated content will take effect in the EU internal market as of 2 August 2026. This regulatory measure is intended to enable consumers to transparently distinguish between original art and automated products (21). The applicability in Liechtenstein results from Art 102 of the EEA Agreement (5). The procedural enforcement of claims, however, remains complicated if the technical infrastructure of the providers is located outside the EEA (22) (23). Notwithstanding this, the user of an AI system is liable for copyright infringements even without knowledge of a reproduction of protected content (the so-called "memory effect") (16).
Conclusion
In conclusion, the intersection of art and AI necessitates a precise differentiation between technical automation and the human act of creation. While the AI Act ensures market transparency, copyright remains a protective right for outputs that are based on a human expression of thought (22).
We would be pleased to advise you on your copyright law issues and are at your disposal for a consultation.
The Authors:
MMag. Thomas Plattner and M.A. HSG Fabian Jenny
List of References
(1) Bundesamt für Kommunikation [BAKOM], Auslegeordnung zur Regulierung von künstlicher Intelligenz – Bericht an den Bundesrat vom 12. Februar 2025, 4, abrufbar unter: https://www.bakom.admin.ch/dam/bakom/de/dokumente/KI/auslegeordnung_regulierung_ki.pdf (zuletzt abgerufen am 16.04.2026).
(2) Europäisches Parlament, Online-Beitrag «Was ist künstliche Intelligenz und wie wird sie genutzt?», 2020, abrufbar unter: https://www.europarl.europa.eu/topics/de/article/20200827STO85804/was-ist-kunstliche-intelligenz-und-wie-wird-sie-genutzt (zuletzt abgerufen am 16.04.2026).
(3) Martina Merker, Die Auswirkungen von KI-generierter Kunst auf den Kunstmarkt und das deutsche Urheberrecht, KUR 2025, S. 89 ff., S. 89.
(4) Martina Merker, KUR 2025, S. 89 ff., S. 89 f.
(5) Stabsstelle für digitale Innovation [SFID], Monitoring EWR-Digitalisierungsrechtsakte, abrufbar unter: https://www.llv.li/de/landesverwaltung/stabsstelle-fuer-digitale-innovation/digitalisierung/monitoring-ewr-digitalisierungsrechtsakte (zuletzt abgerufen am 16.04.2026).
(6) Bundesamt für Kommunikation [BAKOM], Auslegeordnung zur Regulierung von künstlicher Intelligenz – Bericht an den Bundesrat vom 12. Februar 2025, 9, abrufbar unter: https://www.bakom.admin.ch/dam/bakom/de/dokumente/KI/auslegeordnung_regulierung_ki.pdf (zuletzt abgerufen am 16.04.2026).
(7) Lara Burkhalter, Urheberrecht bei KI – Trainingsdaten und Output von KI-Modellen im Licht des schweizerischen Urheberrechts, BFH 2024, 1, abrufbar unter: https://bfh.ch/ipst/ai-copyright (zuletzt abgerufen am 16.04.2026).
(8) Martina Merker, KUR 2025, S. 89 ff., S. 91.
(9) Jacques de Werra / Yaniv Benhamou, Kunst und geistiges Eigentum, in: Mosimann/Renold/Raschèr [Hrsg.], Kultur Kunst Recht, 2. Aufl., Basel 2020, 707.
(10) U.S. Copyright Office, Cancellation Decision re: Zarya of the Dawn vom 21. Februar 2023, abrufbar unter: https://www.copyright.gov/docs/zarya-of-the-dawn.pdf (zuletzt abgerufen am 16.04.2026).
(11) Anna Maria Stein, Generative Artificial Intelligence: US Copyright Office denies registration, 2023, abrufbar unter: https://ipkitten.blogspot.com/2023/03/generative-artificial-intelligence-us.html (zuletzt abgerufen am 16.04.2026).
(12) Julia Wismer, Künstlerische Intelligenz: Zum Werkcharakter von Algorithmen und ihrer Urheberschaft an Kunstwerken, Zürich 2025.
(13) Yaniv Benhamou / Ana Andrijevic, The protection of AI-generated pictures under copyright law, Cheltenham 2022, S. 198 ff.
(14) Cyril Dörfler, Das Schöpferprinzip im Immaterialgüterrecht, Bern 2024, S. 217 ff.
(15) LG München I, Endurteil vom 11. November 2025 – 42 O 14139/24, GRUR 2025, 1917.
(16) Eidgenössisches Institut für Geistiges Eigentum [IGE], FAQ – KI und Urheberrecht vom 23. November 2023, 1, abrufbar unter: https://www.ige.ch/de/urheberrecht/faq-ki-und-urheberrecht (zuletzt abgerufen am 16.04.2026).
(17) Florent Thouvenin, WHITE PAPER Urheberrecht, ITSL 2025, 7.
(18) Martina Merker, KUR 2025, S. 89 ff., S. 91 f.
(19) Florent Thouvenin, WHITE PAPER Urheberrecht, ITSL 2025, 4.
(20) Florent Thouvenin, WHITE PAPER Urheberrecht, ITSL 2025, 5.
(21) Martina Merker, KUR 2025, S. 89 ff., S. 93.
(22) Martina Merker, KUR 2025, S. 89 ff., S. 94.
(23) Daniel Schönberger, Deep Copyright: Up- and Downstream Questions Related to Artificial Intelligence (AI) and Machine Learning (ML), in: Jacques de Werra [Hrsg.], Droit d’auteur 4.0 – Copyright 4.0, Genf 2018, S. 145 ff.
