This feels disingenuous to me. It would be fairly trivial, for example, to store a copy of all keys, encrypted with the government’s public key. Of course, there’s a million eats to go wrong, but that’s different from “mathematically impossible.”
https://www.washingtonpost.com/news/powerpost/paloma/the-cyb...
https://www.schneier.com/blog/archives/2018/05/ray_ozzies_en...
https://www.justsecurity.org/53316/criminalize-security-crim...
As is pointed out in the Schneier article, the problems with a key escrow scheme are on the law enforcement side of things. They could lose access to their keys, especially if a lot of different agencies have keys.
Those are difficulties that can in theory be overcome, although it may not be practical to do so. That's a far cry from a pi = 3 issue.
This is approaching a pi = 3 level falsehood because of the “in no way compromises” clause. There are many schemes that are outright illegal (in my not a lawyer interpretation of this law), and it nakedly makes the other schemes harder with state actors as additional points of failure.
Appealing to Schneier on the topic of encryption is not an irrelevant appeal.
I’m not a cryptographer but I assume there are other schemes that are at least weakened by the requirement of a third party holding a key, much like the TSA master lock program was broken by statistical analysis of locks that were mastered this way.
But the mathematical impossibility if this aside, there is a very real practical impossibility if trusting an organization as large as the US government to keep such a database secure. There are better ways to help law enforcement than blowing such a large gaping hole in the web.