Lets face it. Most, if not all publicly accessible content had been picked already.
Having said that, it seems there had been an update from SC:
“SoundCloud has never used artist content to train AI models, nor do we develop AI tools .....”
SoundCloud seems to be saying that THEY, SoundCloud have never themselves, personally used their catalog to train AI.... But stop short of saying they haven't, or aren't allowing others to do that very thing.
Reading the actual terms of use agreement is rather confusing. In the Grant of license section where it discussed this topic, it seems to say yes they do allow training and in the next section they seem to say the opposite. The end of paragraph 2 into paragraph 3 where it says.... And I quote from the agreement itself:
"You explicitly agree that your Content may be used to inform, train, develop or serve as input to artificial intelligence or machine intelligence technologies or services as part of and for providing the services. For the avoidance of doubt, neither SoundCloud nor any third party is allowed to (i) use, copy or
reproduce any Content delivered to the Platform under separate agreements, which is owned or controlled by third party rights holders (including artwork, images, logos, audio and audiovisual recordings (and any part thereof), underlying musical works and lyrics, and metadata) for the purposes of informing, training developing (or as input to) artificial intelligence technologies without authorization from the applicable rightsholders. You can limit and restrict the availability of certain of your Content to other users of the Platform, and to users of Linked Services, at any time using the permissions tab in the track edit section for each sound you upload, subject to the provisions of the Disclaimer section below."
Sounds like"lawyer speak" to me that hides a loophole big enough to sail an entire US naval battleship fleet through with room to spare.