The NSA has many jobs, and doing public research in crypto is only one of
them -- and a particularly small one at that. For all we know, they had
an attack on MD5 ten years before anyone else and didn't tell anyone
because keeping it secret made it useful for one of their other jobs.
Yes, but they're supposed to look after US _National_ security, not their
own security. Since people in strategic jobs make so much use of hash
functions in crypto, that means it is most certainly an important part of
the NSA's function to ensure that there are good hash functions available.
They've fallen down on that job.
I don't see how that follows.
Because previously, the NSA has done things that it took open researchers
years, even decades, to figure out. But not any more.