LLM Training Fair Use Ruling (US)
Tags: Regulatory
Overview
Bartz v. Anthropic is a notable U.S. federal case in which multiple authors and publishers challenged Anthropic’s use of copyrighted works to train its large language models (LLMs). In a partial decision on the merits, the court found that Anthropic’s use of lawfully obtained copyrighted materials for model training qualified as fair use, while use of pirated materials did not.
Key Takeaways
- Scope limited to training data: The court’s finding was confined to the AI system’s training materials and did not address the legality of any model outputs.
- “Spectacularly transformative”: The court characterized the training process on copyrighted works as highly transformative, analogizing it to humans learning to read, internalizing content, and subsequently creating new works.
- Purpose matters: The court reasoned that the purpose of training was not to reproduce works verbatim, but to generate new, novel text.
- Settlement and procedural posture: The matter settled before resolution of all remaining issues, and the fair‑use decision was not appealed.
Implications for Digital Advertising
- Training vs. outputs: The decision underscores a legal distinction between ingestion of content for training and the generation of outputs—important for advertisers evaluating risk around AI‑assisted creative.
- Lawful sourcing emphasis: Lawful acquisition of training data is critical. Vendors should document data provenance and avoid materials obtained in violation of law or license.
- Contractual risk allocation: Buyers may seek warranties on lawful sourcing and license compliance, while vendors may limit liability related to outputs—a dynamic reflected in contemporary AI terms and licensing agreements.
Context
This case forms part of broader litigation since 2020 against AI developers and coincides with a market shift where many publishers pursue licensing deals with AI companies rather than litigation. The regulatory and judicial landscape remains fluid, necessitating careful due diligence and contractual controls.
Sources:- IAB_AI_Intellectual_Property_and_Transactions_Digital_Advertising_Playbook_December_2025.pdf
← Back to Knowledge Base