The TOS clearly bans the use of licensed materials, such as music for large labels, to train any AI model, including Generative AI. The sound cloud allows the possibility of using other types of materials uploaded to TOS AI.
The important thing is, to date, no such use has been made, and the sound cloud will introduce a strong internal permission control to govern any future potential use. If we ever consider using the USer user content for training of generative AI models, we will be obliged to be less than a clear opt-out mechanism-less than a minimum and transparency with our creative community.