Edit: some people pointed out that whisper would do a good job with transcription but there's other things like tweaking the model which is essentially training it and there is things like building their own summarization systems that may be bespoke by customer. At my work we use some AI that answers HR and other types of questions that are kind of trained on our company specific questions and it actually does a great job but that does mean that we have to allow our data to be used for AI training. We're also using this system to do first tier tech support and some of our developer channels for very common questions and it works great because it finds those common questions and gets an answer before a human's even able to pay attention. Both of those approaches could be enabled by these terms of service changes
This is where zero knowledge federated learning comes in. Unfortunately, this is very much a tomorrow technology (it needs the infrastructure to support it). Why invest in privacy-preserving methods for training machine learning models tomorrow when you can steal users private information today (or even better, bully them into doing so by being the defacto VC that everyone needs to use because of network effects).