A new policy paper argues AI training on public web data should stay legal under fair use, stirring fresh pushback from the music industry.
Google's fair use argument
Google's fair use argument
- The paper says training models on public data counts as transformative use.
- It compares the practice to an art student learning from a gallery.
- Copyright concerns should focus on outputs, not on training inputs.
- Kent Walker published the 21-page paper on Thursday, June 25.
- Google offers machine-readable tags letting sites block AI training.
- The company says it is exploring new partnerships and value-exchange models.
- Google has paid for access to specialized, non-public content.
- Google wants enforcement to target copied outputs, not training methods.
- It opposes automated filters that judge content as too similar.
- The company favors standard notice-and-takedown systems for infringement.
- Google backs the NO FAKES Act, which protects voices and likenesses.
- The RIAA sued Suno and Udio for mass copyright infringement in 2024.
- Udio later signed licensing deals with Universal, Warner, Merlin, and Kobalt.
- Sony Music's case against Udio remains active in court.
- Music publishers sued Anthropic over Claude's use of song lyrics.
- Publishers filed a bigger suit, seeking over $3 billion in damages.
- The RIAA and NMPA urged courts to reject Anthropic's fair-use defense.
- NMPA struck an industry-wide licensing deal with Udio in June.
- David Israelite called it the first deal with a major AI music company.
- Independent artists sued Google in March over its Lyria 3 model.
- They allege Google used YouTube recordings without permission for training.
- Google says artists licensed their music through YouTube's terms of service.
- Google proposes an industry-funded body called FARO to police frontier models.
- The group would set safety standards and verify audits for developers.
- Google compares the idea to regulators like FINRA and NERC.
- Every day AI uses should follow existing laws, the paper argues.
