OpenAI Faces Copyright Ruling in Germany, Signaling Potential Shift for AI and Art
A German court recently ruled against OpenAI, finding the company liable for copyright infringement related to the use of song lyrics in the training data for ChatGPT. The court resolute that OpenAI infringed on copyright by making the lyrics available to the public through shareable links generated by the AI.
The ruling mandates that OpenAI disclose how frequently specific lyrics were used in its training data, when (if ever) profit was generated from their use, and requires the company to cease storing and reproducing the lyrics. Monetary damages will be assessed at a later date.
This decision contrasts with a recent UK court case where Getty Images lost its claim against Stability AI. The UK judge ruled that because stability AI’s model doesn’t store or reproduce copyrighted works, it doesn’t create an “infringing copy.” Tho, legal scholar Andres Guadamuz notes the German court’s reliance on research concerning machine “memorization,” a process more readily demonstrated with text like lyrics than with images.
The German court’s decision aligns with established intellectual property law, asserting that if copyrighted material can be reproduced perceptibly, it constitutes a monetizable copy - a principle applicable to both physical media like CDs and within Large language Models (LLMs).
Guadamuz does raise concerns about the ruling’s attempt to establish training without memorization as the legal standard, citing a misinterpretation of EU data-mining law.He argues the ruling implies memorization always occurs during training, which isn’t necessarily true. This potential legal ambiguity could impact how companies interpret the case moving forward. However, Guadamuz predicts the ruling will likely lead to the progress of a licensing market for copyrighted material used in AI training.
The situation echoes the evolving response to OpenAI’s Sora 2, where initial concerns about copyright and likeness from actors and copyright holders have gradually shifted towards exploring revenue-sharing models. This trend is visible in recent partnerships between major U.S. record labels, like Global Music Group and Udio, with companies they previously sued.
Despite this movement towards collaboration with powerful stakeholders, the benefits for smaller artists, writers, and creators – those concerned about their work becoming obsolete in the age of generative AI – remain unclear. The ruling highlights the ongoing challenge of ensuring equitable compensation and recognition for all copyright holders in the rapidly evolving landscape of AI-generated content.
