Generative AI Training Fair Use Limits
Tags: Case Studies, Regulatory
TL;DR
- Delaware District Court found Westlaw’s headnotes sufficiently creative to be copyrightable.
- Creating temporary copies to train an AI used in a competing product was not fair use.
- Signals greater IP risk for training on copyrighted corpora without licenses.
Why it matters for HK marketers: Check whether your vendors train models on licensed data—IP exposure can cascade to your workflows and content.
Decision highlights
- Protectable content: Westlaw headnotes deemed creative, not mere facts.
- Training copies: Non‑transformative use to build a competing tool weighed against fair use.
Marketing relevance
- Generative tools in ad production often rely on large training sets that may include copyrighted material.
- Contracts and indemnities around training data matter.
So what for marketers
Ask vendors how training data is sourced and whether your data is used for “system improvements.” Tighten indemnification and opt‑out of cross‑customer training where possible.
← Back to Knowledge Base