A kleptographic vulnerability (one which cannot be exploited without breaking a cryptographically strong problem) is entirely different from the plain "ECC believed strong but actually weak" situation. Once the latter vulnerability is known anyone could break it, but the former vulnerability is strong even if the theoretical basis for the backdoor is discovered.
It's the same fundamental difference between building a computational-theoretical hard algorithm and security-by-obscurity. The NSA is not dumb enough to bet on their trick remaining obscure - that stuff inevitably gets rediscovered, whether it's 5 years or 20 years. But it's easy to precompute E in the DUAL_EC_DRBN algorithm, while it's cryptographically strong to try and reverse it after the fact. Trapdoor functions work like one-way secure hashes (eg SHA) by design - easy in forwards, computationally infeasible (as a design goal) in reverse.
The patent on the concept of using that as a backdoor (although not in that exact phrasing) was filed in Jan 2005 [1] (years before the selection of the NIST curves in FIPS 186-3) and it's reasonable to believe the NSA would know of the existence of the backdoor since the filing at a minimum - if not before. It was published for public comment in 2007 [1] but didn't gain much publicity until granted in 2013(!) [2] after Snowden had gotten the disclosure train rolling.
If you have a specific page/paragraph reference you'd like to cite: please do, it's a topic I feel strongly about since the consequences of a compromised kleptographic backdoor could be rather extreme for anyone who uses those curves.
What would it mean for a curve to have a "kleptographic" back door? Only one base point P is defined in the curve parameters, so there is no hidden relationship to take advantage of. It is possible there are weaknesses in the NIST curves, but if so:
1. They must lie in the curve parameters themselves, i.e. something anyone could conceivably discover.
2. They must rely on a significant advance in ECDLP, e.g. a new class of weak curve.
What if the NIST curves were backdoored not by the NSA as a whole, but a rogue individual within the NSA, with the goal of making that backdoor available to a foreign power and/or the highest bidder?
It seems unlikely to me, because the NSA has so many brilliant people working for them, but it also seems like (at least superficially) it would explain just about every element of their reaction to this situation. They're used to being years or decades ahead of everyone else (e.g. differential cryptanalysis), and being caught off-guard would be an uncomfortable position for them.
</tinfoilhat>
You could argue that that is indeed what the announcement was doing. It was unrelated to P-256 being dropped.
(P-256 being dropped does point to the whole "it's not as hard as we thought it was" theory)
I suspect this falls into the "too good to be true" category.
But then you can't actually use it to do anything and its existence is purely a downside. And you can't put it online - in any fashion - without exposing it to some possibility of compromise. Whether that's by one-or-more sysadmins who take a massive bribe (compromising two decades of SCI-level USG secrets => name a number and we'll pay in Billions), or users who can perform sidechannel attacks, and so on.
Everyone knows what happens once you expose real systems to real users/admins (should be read as "attackers"). You cannot design a black box that miraculously requires no administration and is not vulnerable to any sidechannel attack when your threat category is state-level actors.
I actually suspect the NSA does fall into the category of "altruistic actors" here. Their number-one concern is making sure that nobody can decrypt the things that nobody should be reading. I don't think they're above pulling a kleptographic backdoor if there's a computationally-strong guarantee that nobody else can break it (perhaps while learning a hard lesson about socially-strong security) but at the end of the day I do trust that they won't be broadcasting State Secrets in something that will be publically accessible within a decade.
When dealing with a state actor like the NSA you need to reverse Hanlon's Razor - they may be malicious, but they're not stupid. They know when the jig is up and it's time to cut their losses.
The problem with this theory is that in the event of compromise you'd expect the NSA to deprecate all of the curves, and not just P-256. It is also possible that this backdoor is not a magic bullet, and it merely makes an attack computationally feasible instead of decrypting it outright.
It does smell a lot more like there's been an advance in analysis, whether classical or quantum.