alt Hacker NewsThis all loops back to the same thing:
you _really_ shouldn't be allowed to train on information without having a copyright license explicitly allowing it
"publicly available" isn't the same as "anyone can do whatever they want with it", just anyone can read it/use it for research
In many cases government texts are not covered by copyright, so it may not even be relevant here, regardless of it is is allowed to copy the data or not.