The AI Act will be fully applicable from 2 August 2026.
Providers of GPAI models must respect Text and Data Mining (TDM) opt-outs.
2.1 Legal Basis: Article 53(1)(c) AI Act and Directive (EU) 2019/790
The Copyright Chapter of the Code directly addresses one of the most contentious legal questions in AI governance: the use of copyrighted material in training GPAI models and the risk of infringing outputs. Article 53(1)(c) AI Act requires GPAI providers to “identify and respect copyright protection and rights reservations” within their datasets.
+
This obligation complements the framework of Directive (EU) 2019/790 on copyright and related rights in the Digital Single Market (DSM Directive). Notably, Article 4(3) DSM Directive allows rightsholders to exclude their works from text and data mining (TDM) operations via machine-readable opt-outs.
Well, in the eyes of the EU, the entire 'fair use' thing doesn't exist there at all (per the EU's own JURI, which I think is roughly the equivalent of the U.S.'s Office of the Attorney General, with similar duties around defining the canonical interpretations of the laws).
Two relevant bits, dug out from the 175-page whole:
> Although the Act tries to
address this by extending obligations to any provider placing a GPAI model on the EU market, the
extraterritorial enforcement of these obligations remains highly uncertain due to the territorial nature
of copyright law and the practical difficulty of pursuing infringement claims when training occurs under
foreign legal standards, such as U.S. fair use.
and:
> Finally, it is important to clarify that the current EU framework provides a closed list of exceptions and
does not recognise a general fair use defence. As a result, AI-generated outputs that include protected
expression without a valid exception remain unlawful.
It seems to be dated the same month as DDG's analysis, July 2025, so I would expect the MIT Non-AI License that we're discussing here to be much more defensible in the EU than in the U.S. — as long as one studies that full 175-page "Generative AI and Copyright" analysis and ensures that it addresses the salient points necessary to apply and enforce in EU copyright terms. (Queued for my someday-future :)
From a U.S. standpoint: Licensing is a function of copyright. A work not subject to copyright cannot be licensed productively as-is, as the public domain quality of the work is a trivial and conclusive defense against a licensor’s claims of copyright violation. Fair use is not subject to copyright. Since licensing enforcement is only possible with an applicable copyright, enforcement cannot be completed against fair uses, as copyright law is not applicable to fair uses and therefore licensing enforcement has no legal basis. However, a judgment may overturn a defense of fair use brought against a defendant in a licensing enforcement claim, which would then subject the defendant’s use to copyright law and thus to license enforcement.
It’s midnight now, so you’re on your own to dig up and review specific instances of relevant case law, or to contrast with non-U.S. laws. Licensing above refers to i.e. LICENSE files of the specific sort that this post is about ("MIT Non-AI License"); other definitions of licensing, as well as e.g. DMCA exceptions, exist that might be of interest for you to explore further. I believe there’s been a handful of cases related to AI and fair use this past year, but as with all such defenses, unique circumstances are common enough that I hesitate to suggest any future outcome as 100% certain without much more case law than AI has today. (I am not your lawyer, this is not legal advice.)
