Do generative AI models contain “copies” of the works on which they are trained?
This question lies at the centre of current policy debates on copyright, innovation, and the regulation of AI. It has recently been tested in two high-profile European disputes that appear to point in different directions:
- GEMA v OpenAI (Case 42 O 14139/24, Munich I Regional Court, 11 November 2025) – examining whether AI outputs and internal model memory may constitute copyright infringement
- Getty Images (US) Inc & Ors v Stability AI Ltd [2025] EWHC 2863 (Ch) – reflecting a contrasting judicial approach in the UK
The stakes could hardly be higher. If the reasoning adopted by the Munich I Regional Court in GEMA gains traction (a position also explored in Professor Shemtov’s co-authored Heart of the Matter paper, September 2024), it could significantly destabilise the current EU copyright settlement. In particular, Articles 3 and 4 of the DSM Directive, designed to facilitate text and data mining, may offer limited reassurance if liability is instead grounded in a finding that AI models themselves embody copies of protected works. The implications for innovation policy, market entry, and the governance of AI development would be far-reaching.
It is against this wider set of structural challenges that Professor Shemtov launched the new specialist LLM module IP Law in the AI Era at Queen Mary University of London. The module interrogates how AI technologies are exerting pressure across the intellectual property system as a whole, from copyright and patents to trade secrets and trade marks. Recent cases such as GEMA and Getty serve as concrete illustrations of the kinds of regulatory tensions now confronting policymakers, courts, and industry alike.
This webinar provides a timely forum to reflect on these developments and their implications for the future design of IP law in an AI-driven economy.
