Whalesbook
The Munich Regional Court has ruled that OpenAI's ChatGPT infringed copyright by memorizing and reproducing German song lyrics. The court found in favor of GEMA, a music rights organization, stating that the AI models' ability to 'regurgitate' lyrics constituted infringement during training and output. OpenAI was ordered to pay damages and cease infringing activities.
The Munich Regional Court I has issued a significant ruling in Gema v. OpenAI, finding that OpenAI's ChatGPT infringed copyright by storing and reproducing song lyrics. The court largely sided with GEMA, a German music rights organization, on claims involving nine German song lyrics.
The lawsuit was filed against two entities of the OpenAI group, alleging unauthorized use of nine German song lyrics, including works by Herbert Grönemeyer. GEMA argued that the lyrics were reproduced within ChatGPT's large language models (LLMs) during the training phase and then publicly communicated when the chatbot generated them in response to user prompts.
OpenAI contended that its models learn statistical patterns, do not store specific data, and thus don't create copyright-protected copies. They also invoked the Text and Data Mining (TDM) exception and argued that end-users, not the platform, should be liable for generated content.
The court found that the AI models' ability to 'regurgitate' lyrics verbatim demonstrated reproduction. It ruled that memorization as numerical probability values still constitutes reproduction under copyright law. The TDM exception was deemed inapplicable, as it allows copies only for analysis, not for long-term memorization and reproduction of entire works, which infringes exploitation rights. The court also held OpenAI directly liable for the public communication of the lyrics, stating simple prompts do not shift liability to the user.
OpenAI was ordered to pay GEMA damages of €4,620.70 and to cease infringing activities. The court found OpenAI negligent, aware of the memorization risks since at least 2021, rejecting their requests for a grace period.
Impact
This ruling sets a precedent for AI copyright infringement cases, particularly concerning training data and outputs. It could lead to increased scrutiny and potential litigation for AI developers globally, influencing how LLMs are trained and used. Investors in AI and technology companies may need to assess potential liabilities and regulatory risks.
Rating: 7/10
Difficult Terms:
Large Language Models (LLMs): Advanced AI programs trained on vast amounts of text data, capable of understanding and generating human-like text.
Copyright: The exclusive legal right granted to the creator of original works to control their use and distribution.
GEMA: The Society for musical performing and mechanical reproduction rights in Germany, a music rights organization.
Text and Data Mining (TDM): A method used in AI to analyze large datasets of text and data to identify patterns and extract information.
Infringement: The violation of a law or right, in this case, copyright law.
Reproduction: The act of making a copy of a work.
Public Communication: Making a work available to the public, such as through broadcasting or online display.
Negligence: Failure to exercise the care that a reasonably prudent person would exercise in like circumstances.
