For the record: (a) I can't find anything Schneier has ever published on ECC; it is a notable omission in his most recent crypto book (Cryptography Engineering, (b) the ECC issue is confused by Dual-EC-DRBG, the ECC-derived CSPRNG that nobody uses but is now thought to be a deliberately weak NSA design, (c) his reasoning for avoiding ECC (the constants are suspect) is a little bit of a stretch given that the most popular curves have a relatively straightforward derivation, and (d) it's downright weird to point a finger at all of elliptic curve cryptography based on a single set of constants; surely he's not implicating the Edwards curves Bernstein and Lange have been promoting, for instance.

The recommendation is confused enough that I'm inclined to dismiss it.

The deterministically derived ECC constants are derived by seeding a hash function with an extremely high entropy input (> 100 bits), and taking the first usable result. This is effectively the same as choosing the parameter. The NSA had freedom to specify this very high entropy seed value, and could have done so by iteratively trying seed values until they got a curve that weak using techniques only known to the NSA; it's just a way to disguise the origin of the curve by making it sound random. The situation with Kobolitz curves is only slightly better.

Check the date -- the proposal is a month old. So Bruce Schneier's comments may be relevant, but one can't really accuse Brian Smith of ignoring them.

Good point, I missed that on the first pass, I look forward to an update from him or a comment on this from Schneier.