The cryptography community seems to already be aware of this risk. I've read about random number algorithms that mix the results of many randomness sources so that one compromised source cannot compromise the result.
> and the next-best targets would be "enclaves"
Yeah, we should watch out for this. Secure cryptoprocessors are a great thing but we need to be able to trust them. We already have hardware that runs free software but the chip itself can of course be compromised...
> then back-dooring the algorithms themselves
I trust that the worldwide cryptography community will never let that happen. From what I've read they no longer trust the NSA implicitly.