Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

AI Memorisation and the Reproduction Right: Protected Works in Generative AI Models under EU Copyright Law

Nguyen, Le Thao Vi LU (2026) JAEM03 20261
Faculty of Law
Department of Law
Abstract
This thesis examines whether memorisation in Generative AI models, particularly large language models, can constitute an infringement of the reproduction right under EU copyright law. It focuses on situations where protected expression is not only used as training data or reproduced in outputs, but may also be retained within a trained model itself.

The thesis adopts a doctrinal legal method, supported by selected technical literature on AI memorisation. It first explains how memorisation differs from general learning, statistical abstraction, extraction, regurgitation and ordinary output similarity. It then analyses the reproduction right under Article 2 of the InfoSoc Directive, with particular attention to its broad and... (More)
This thesis examines whether memorisation in Generative AI models, particularly large language models, can constitute an infringement of the reproduction right under EU copyright law. It focuses on situations where protected expression is not only used as training data or reproduced in outputs, but may also be retained within a trained model itself.

The thesis adopts a doctrinal legal method, supported by selected technical literature on AI memorisation. It first explains how memorisation differs from general learning, statistical abstraction, extraction, regurgitation and ordinary output similarity. It then analyses the reproduction right under Article 2 of the InfoSoc Directive, with particular attention to its broad and technologically neutral scope. The thesis argues that AI memorisation may amount to reproduction where identifiable protected expression from a specific work is retained at model level and can be reconstructed, extracted or regurgitated with sufficient reliability.

The thesis further considers whether existing EU copyright exceptions and limitations, including temporary copying, incidental inclusion and text and data mining, can apply to such memorisation. It finds that these exceptions provide only limited answers, especially where protected expression remains retained in a trained model and capable of later reproduction. The thesis concludes that AI memorisation can, in specific circumstances, constitute an unauthorised reproduction and therefore an infringement of the reproduction right, unless a valid exception or limitation applies. It also shows that the current EU copyright framework still leaves uncertainty for rightholders, AI providers and future regulation. (Less)
Please use this url to cite or link to this publication:
author
Nguyen, Le Thao Vi LU
supervisor
organization
course
JAEM03 20261
year
type
H2 - Master's Degree (Two Years)
subject
keywords
Copyright, Reproduction right, AI
language
English
id
9232152
date added to LUP
2026-06-10 11:09:28
date last changed
2026-06-10 11:09:28
@misc{9232152,
  abstract     = {{This thesis examines whether memorisation in Generative AI models, particularly large language models, can constitute an infringement of the reproduction right under EU copyright law. It focuses on situations where protected expression is not only used as training data or reproduced in outputs, but may also be retained within a trained model itself.

The thesis adopts a doctrinal legal method, supported by selected technical literature on AI memorisation. It first explains how memorisation differs from general learning, statistical abstraction, extraction, regurgitation and ordinary output similarity. It then analyses the reproduction right under Article 2 of the InfoSoc Directive, with particular attention to its broad and technologically neutral scope. The thesis argues that AI memorisation may amount to reproduction where identifiable protected expression from a specific work is retained at model level and can be reconstructed, extracted or regurgitated with sufficient reliability.

The thesis further considers whether existing EU copyright exceptions and limitations, including temporary copying, incidental inclusion and text and data mining, can apply to such memorisation. It finds that these exceptions provide only limited answers, especially where protected expression remains retained in a trained model and capable of later reproduction. The thesis concludes that AI memorisation can, in specific circumstances, constitute an unauthorised reproduction and therefore an infringement of the reproduction right, unless a valid exception or limitation applies. It also shows that the current EU copyright framework still leaves uncertainty for rightholders, AI providers and future regulation.}},
  author       = {{Nguyen, Le Thao Vi}},
  language     = {{eng}},
  note         = {{Student Paper}},
  title        = {{AI Memorisation and the Reproduction Right: Protected Works in Generative AI Models under EU Copyright Law}},
  year         = {{2026}},
}