When the Client Trained a Model on Someone Else’s Data, Copyright Exposure and Compliance

This session examines the defense of copyright infringement and data-scraping claims arising from AI model training, focusing on the three landmark 2025 federal district court rulings that now define the fair use landscape for LLM training data. Attorneys will learn how to evaluate client exposure based on data provenance, market substitution evidence, and output containment, and how to deploy preemption, early motion practice, and settlement benchmarks in active litigation. Participants will leave with a structured defense framework applicable to both copyright and state-law scraping claims.

This session covers the compliance and transactional obligations attorneys must understand when a client trains or deploys a generative AI model using third-party data. Attendees will examine California’s AB 2013 Training Data Transparency Act — its 12 mandatory disclosure categories, retroactive scope, and early compliance patterns — alongside the EU AI Act’s parallel GPAI training data obligations. The session also addresses practical training data due diligence frameworks, the structure of content licensing deals, and how the emerging licensing market intersects with fair use defenses in ongoing litigation.