GEMA v OpenAI: Where Will The Balance Between Tech Platforms And Content Creators Fall?

Since the release of ChatGPT in late 2022, copyright and technology lawyers have been debating the applicability of traditional copyright law to the development of the generative artificial intelligence (GenAI) tools that have flooded the technology market. In a recent German case, copyright experts have their first definitive judicial decision that finds that GenAI has created a copy that infringes existing copyright works. We look at what the case says and discuss what it means for balancing the rights of existing content creators with those of the technology companies of the AI future.

Background to the GEMA v OpenAI case

GEMA (Gesellschaft für musikalische Aufführungs- und mechanische Vervielfältigungsrechte),[1] a German collecting society which (like APRA AMCOS in Australia) represents composers, lyricists and music publishers by licensing and collecting royalties for public performance and mechanical reproduction of musical works on behalf of their members, claimed that OpenAI was using the lyrics of famous songs of its members for AI training purposes in an unlawful manner.  Test results had shown that OpenAI tools like ChatGPT were reproducing such lyrics verbatim within their outputs. GEMA claimed that this behaviour constitutes an infringement of its members’ copyright in such compositions and lyrics.

What did the court decide?

On 11 November 2025, the Regional Court in Munich (LG München I, Urt. 42 O 14139/24 v. 11.11.2025) upheld GEMA’s claim in most respects.  The court found:

• that the large language model (LLM) must have included a reproduction (“memorisation”) of protected works in the Sense of § 16 UrhG (Urheberrechtsgesetz – the German Copyright Act[2]) as lyrics were literally reproduced verbatim within its output – which then also constitutes an infringement of copyright holders’ right to make such content publicly available under § 19a UrhG;
• the exception for “Text and Data Mining” under § 44b UrhG does not apply to the copies as this section may only cover mere data extraction for AI training purposes, not the literal reproduction of protected works; and
• finally, the court did not see any evidence supporting an infringement of moral rights under § 13 and 14 UrhG which had been claimed by GEMA due to the fact that altered lyrics may become falsely attributed to affected artists.

The decision of the court is therefore heavily in favour of copyright holders.

In contrast, the Regional Court in Hamburg[3] took a different approach in September 2025, which has recently been confirmed by the local Higher Regional Court.[4] The outcome of that case is less supportive of creators and is less restrictive for developers of AI systems.

In that case, photographs have been used for AI training purposes to verify whether a picture’s description actually matches the respective picture. The Hamburg courts found this usage fell within both by the text and data mining (TDM) exception within § 44b UrhG as well as the exception for scientific research provided by § 60d UrhG. Importantly, the courts deemed an AI usage disclaimer provided by the rights holder insufficient because it had not been provided in “machine readable form”.

Therefore, we can conclude that the underlying cases vary significantly in their treatment of original word-by-word lyrics and image-to-text relationships. It may be argued that the Hamburg courts interpreted the respective provisions much more in favour of AI service providers – in contrast to the approach of Munich’s Regional Court which admittedly had to deal with the practically far more important aspect of copyrighted content being actually reproduced in the output of a GenAI tool.

What does this imply for AI service providers?

The GEMA ruling strengthens the rights of copyright holders against LLMs that generate (or potentially contain) reproductions of their content. The court makes clear that in its opinion the described utilisation of protected content is not covered by statutory copyright exceptions – and thus requires other legal justification such as the particular copyright holders’ consent.

It is obvious that the GEMA decision will be challenged before all competent national and in all likelihood European courts. This may leave the position in Germany and the wider European Union uncertain for some time.  If the decision stands, it will impose a significant hurdle for LLM providers and their business models where availability of large amounts of training data from sources protected by copyright is crucial. It would become much more difficult to acquire such data when the consent of affected content creators is deemed to be required.  This would also be likely to significantly affect commercial framework conditions of connected AI services.

It will be interesting to see whether this assessment stands throughout the likely appeals process – and whether the EU’s TDM exception for training purposes may be interpreted more widely to facilitate development of innovative AI-driven services in the future.

Global Impact

This case also advances the debate in Australia,[5] the UK[6] and other jurisdictions on whether a new TDM exception from copyright infringement is needed to train LLMs.  It raises doubt for LLM operators as to whether an exception from data ingest for training is enough, if the output can still be found to infringe the works used for training.  Indeed, Australia is already considering the viability of a licensing regime for copyright works, as being consulted on by the Copyright and Artificial Intelligence Reference Group (CAIRG).[7]

Many cases continue around the world on the implications of GenAI for copyright holders – most notably the New York Times case against OpenAI filed in December 2023[8] and the Canadian news organisations equivalent case in Ontario.[9]  In the UK, Getty Images has had their secondary infringement of copyright claims rejected but has established that the reproduction of their trade mark watermark in GenAI output amounted to trade mark infringement[10] – a timely reminder of the complexity of copyright infringement claims relative to claims of trade mark infringement.

Given the domicile of most LLM operators and the scale of the relevant companies, the New York Times case remains pivotal to the practical outcome of this area of the law.  Without legislative reform or a unequivocal judicial ruling, it is hard to see how LLM operators can have sufficient legal certainty to fully exploit their technology without carrying the ongoing risk of intellectual property claims.  Furthermore for LLM customers, if the extent of works used to train a model is not known, those using the output of such GenAI do not know the full gamut of potential claimants whose works they may be held to infringe – that is, a user could wonder whether the website that ChatGPT wrote for them infringes the latest Richard Osman novel and whether they could be exposed to a claim from Penguin Random House for such infringement.

As different jurisdictions interpret their existing laws or introduce TDM exceptions, there is likely to be further divergence in rules between jurisdictions encouraging LLM operators to engage in forum shopping and smaller or entrepreneurial countries to offer themselves as “data mining havens” to host data centres and LLM development.

This article was first published in CAMLA’s Communications Law Bulletin v44.4 December 2025.

[1] https://www.gema.de/en/w/help/gema/organisation/questions-and-answers/what-is-gema

[2] Gesetz über Urheberrecht und verwandte Schutzrechte, https://www.gesetze-im-internet.de/urhg/.

[3] Ruling as of 27 September 2025 – 310 O 227/23.

[4] Ruling as of 10 December 2025 – 5 U 104/24.

[5] The Hon Michelle Rowland MP (Commonwealth Attorney-General) Albanese Government to ensure Australia is prepared for future copyright challenges emerging from AI (26 October 2025) https://ministers.ag.gov.au/media-centre/albanese-government-ensure-australia-prepared-future-copyright-challenges-emerging-ai-26-10-2025.

[6] Department for Science, Innovation and Technology,  Department for Culture, Media and Sport ‘Copyright and artificial intelligence statement of progress under Section 137 Data (Use and Access) Act’ (15 December 2025) https://www.gov.uk/government/publications/copyright-and-artificial-intelligence-progress-report/copyright-and-artificial-intelligence-statement-of-progress-under-section-137-data-use-and-access-act.

[7] Attorney-General’s Department Copyright and Artificial Intelligence Reference Group (CAIRG) https://www.ag.gov.au/rights-and-protections/copyright/copyright-and-artificial-intelligence-reference-group-cairg (accessed 19 December 2025).

[8] Grynbaum, MM and Mac, R, ‘The Times Sues OpenAI and Microsoft Over A.I. Use of Copyrighted Work’ The New York Times (27 December 2023) https://www.nytimes.com/2023/12/27/business/media/new-york-times-open-ai-microsoft-lawsuit.html.

[9] Karadeglija, A, ‘Lawsuit by Canadian news publishers against OpenAI gets green light to proceed in Ontario’ Canadian Broadcasting Corporation (27 November 2025) https://www.cbc.ca/news/canada/toronto/openai-new-publishers-lawsuit-ontario-9.6995520.

[10] Getty Images (US) Inc and others v Stability AI Ltd [2025] EWHC 2863 (Ch) (UK).