Definition
Training data copyright covers legal disputes over using copyrighted works to train LLMs — fair use, piracy, licensing, and enterprise due diligence.
Key Points
- 2026-09: Authors Guild summary judgment motion alleges intentional LibGen piracy by openai/microsoft (2026-09-27-authors-guild-microsoft-openai-piracy-briefs)
- Internal Slack: “sketchy russian website” optics concern (July 2019)
- Jack Clark acknowledged GPT could “make people unemployed”
- Courts have held obtaining “for free” what one would “ordinarily have to buy” is not fair use (plaintiff argument)
- Enterprise AI projects require training data provenance audits