Fuck the FIPS process; if you made a decent design which was actually useful, a lot of non-FIPS-requiring entities would benefit from it. Design-to-meet, but let only those who actually need FIPS go through the process of assembling and certifying their particular instantiations of the open design for FIPS. For personal use, I'd consider an open design which never touches NIST to be more trustworthy than SafeNet or Thales products.
AND
"Large Internet companies use dedicated hardware to scramble traffic before it is sent. In 2013, the agency planned to be able to decode traffic that was encoded by one of these two encryption chips, either by working with the manufacturers of the chips to insert back doors or by exploiting a security flaw in the chips' design."
(Oh, so what's redacted are the names of two companies which sell this kind of hardware...)
from: http://www.nytimes.com/interactive/2013/09/05/us/documents-r...
All of the big guys use HSMs to protect key material from intruders or malicious insiders. Key generation happens inside a sealed box from which the keys can never leave, except into another identically-configured sealed box from the same vendor. As such, there's no way to inspect the keys being generated - you have to trust the FIPS certification that it's being done correctly.
If--perhaps as part of the FIPS certification process--the NSA were to compromise the key generation function of a handful of popular HSM models, in one fell swoop they would have compromised all of the CAs, DNSSEC, Google, Microsoft, Facebook, Amazon, etc. (Not to mention all of the banks, credit card companies, etc.)
Further data point: in my experience talking to SafeNet at least, they employ a lot of ex-DoD folks who probably still have connections to their old bosses.
This is just speculation, but seems a likely attack vector.