For everyone concerned about the AI systems being trained on copyrighted material: this was always the end-game of that argument. Once the technology was proven out to be useful, someone with a huge IP portfolio was going to slam that portfolio directly into the training data to get their own copyright-unencumbered AI.