Why do you think you've prevented third party supply train risk when you're training on a 30+ Terabyte blob that you have little understanding of its contents.

Why do you think you've prevented third party supply train risk when you're training on a 30+ Terabyte blob that you have little understanding of its contents. It's essentially entirely composed of unfiltered user input.