Google argued that training AI models on publicly available web data should remain protected as fair use in a 21-page policy paper, contending copyright concerns are best addressed at the level of outputs rather than how a model was trained. The paper, written by global affairs president Kent Walker, likens training to an art student taking inspiration from a gallery, and says enforcement should target whether a specific output actually copies an existing work, handled through standard notice-and-takedown rather than automated similarity filters. Google has a direct stake, since independent artists sued it in March over training its Lyria 3 music model on YouTube recordings, a case Google moved to dismiss by arguing the artists licensed their music through YouTube’s terms. The paper’s central proposal is FARO, an independent industry-funded body to set safety standards for frontier models.
