- Recent cases
1.1 The Midjouney case
The US copyright panorama has recently been shaken by the lawsuit filed in June by Disney and Universal against an AI image-generating company, Midjourney[1]; in particular, they argued that there has been unauthorised use of their copyrighted works and, therefore, copyright infringement.
What is noteworthy in this lawsuit is that the plaintiffs refrain from addressing the complex issue of proving that Midjourney has trained its AI with copyrighted images; instead, they point out that the outputs generated are identical to original works, and they are made easily available to the platform’s users.
In addition to direct infringement, the plaintiffs also claim that Midjourney is secondarily liable for the infringing conduct of its users, who generate protected content via prompts. In particular, the studios argue that Midjourney has clear knowledge of copyright infringement, and it has also designed its platform to encourage and induce[2] the reproduction of copyrighted works, as can be seen from its “Explore” page. Additionally, it has failed to block infringing prompts and to implement content filtering through guardrails.
1.2 The OpenAI case
In the meantime, in April, there has been a turning point in a previous lawsuit – The New York Times vs. OpenAI and Microsoft, filed in December 2023[3]. The New York Times had alleged that OpenAI and its partner Microsoft infringed on its copyrights by using millions of its articles to train GPT-based AI systems without permission or compensation.
More specifically, not only is OpenAI supposed to allow the use and reproduction of protected works during the training of its models, but also those models can memorise the works and can generate near-verbatim reproductions, allowing ChatGPT users to read the articles’ contents through summaries or copies without paying the NYT subscription.
The appointed judge has recently delivered an opinion that partially validates the media’s theory and basically affirms the applicability of copyright infringement to generative AI tools, especially when AI is trained on protected content and used to produce and reproduce infringing works.
In particular, the judge has rejected parts of OpenAI and Microsoft’s motion to dismiss, considering the numerous and “widely publicised” examples of copyright infringements named in the complaint. The suit is now set to trial, but the harm to the integrity of copyright management information (CMI)[4] claimed by the NYT has still to be proved.
- The US policy developments
In parallel with such cases, the US Copyright Office (USCO) has focused its attention on the matter, and it has been particularly productive.
In February, the Copyright Office issued the report “Identifying the Economic Implications of Artificial Intelligence for Copyright Policy”[5], produced by a group of scholars discussing the economic issues at the intersection of artificial intelligence and copyright policy.
The report has the merit of elaborating on the possible access regimes for AI developers who need data to train generative AI models, among which the following are noteworthy:
- “Grant Retrospective Conditional Amnesty and Enforce Access”: the solution proposed would give AI developers amnesty for the period in which there was no legal precedent, but it would impose strict restrictions on them for the future. Nevertheless, an author’s compensation fund could be created to partially satisfy their request.
- “Create a New Statutory Blanket License”: under this proposal, Congress should create a statutory blanket license to replace individually negotiated licenses with a collective standard fee for the use of copyrighted works for AI training.
More recently, in May, the Copyright Office released the third issue of the series “Copyright and Artificial Intelligence”[6] regarding Generative AI Training.
The report clarified that creating and deploying a generative AI system[7] using copyright-protected material, with no license or applicable defence, may be considered infringement under the US Copyright Act.
Nevertheless, AI companies could rely on the fair use doctrine, now codified in Section 107 of the Copyright Act. Among the list of four non-exclusive factors that have to be assessed to apply fair use, “the amount and substantiality of the portion used in relation to the copyrighted work as a whole” is the most interesting to analyse.
The Copyright Office noted that the amount made available to the users through outputs is important because, even if a training dataset has incorporated entire copyrighted works, the existence of guardrails preventing the risk of releasing protected content to users could make fair use applicable.
- Conclusions
Summing up the considerations made, a few key points emerge.
Firstly, creating guardrails to safeguard the copyrighted works used to train generative AI could reduce the prejudice caused to rightsholders. Indeed, Disney and Universal expressly asked Midjourney to implement meaningful guardrails in the prompts’ system; nonetheless, the AI company did not respond to the multiple cease-and-desist letters.
Secondly, the development of licensing solutions is central to the matter. In particular, the feasibility of voluntary or compulsory licensing is being discussed while some doubts about scalability, transaction costs and identification of rightsholders remain. The US Copyright Office – in the abovementioned report – underlined the importance of keeping the dialogue on licensing standards and transparency mechanisms alive. Concurrently, on the European side, authors called for transparency and remuneration in AI rules, urging the EU Parliament to act proactively.
In conclusion, as happens very often, a balancing of the opposing rights and interests at stake is desirable to avoid penalisation for AI developers’ initiatives and to safeguard protected works by properly recognising and rewarding the copyright holders.
[1] DISNEY ENTERPRISES, INC., MARVEL CHARACTERS, INC., VL FILM FINANCE LLC, LUCASFILM LTD. LLC, TWENTIETH CENTURY FOX FILM CORPORATION, UNIVERSAL CITY STUDIOS PRODUCTIONS LLLP, and DREAMWORKS ANIMATION L.L.C., v. MIDJOURNEY, INC.; Case 2:25-cv-05275 filed on 6th June 2025.
[2] The studios’ allegation is based on the theory developed in the Supreme Court case. 545 U.S. 913 (2005), METRO-GOLDWYN-MAYER STUDIOS INC. v. GROKSTER, LTD.
[3] THE NEW YORK TIMES COMPANY v. MICROSOFT CORPORATION, OPENAI, INC., OPENAI LP, OPENAI GP, LLC, OPENAI, LLC, OPENAI OPCO LLC, OPENAI GLOBAL LLC, OAI CORPORATION, LLC, and OPENAI HOLDINGS, LLC; Case 1:23-cv-11195 filed on 27th December 2023.
[4] Copyright management information (CMI) is defined in §1202(c) of the Digital Millennium Copyright Act (DMCA) to encompass a work’s title, the identity of the author, and terms and conditions for use of the work.
[5] U.S. Copyright Office, February 12, 2025, “Identifying the Economic Implications of Artificial Intelligence for Copyright Policy”, available here.
[6] U.S. Copyright Office, May 2025, “Part 3: Generative AI Training, pre-publication version”, available here.
[7] The actions that may constitute infringement are data collection and curation; training; retrieval-augmented generation (RAG).
