For Hacker News, the far more important thing to point out is that no matter what block cipher you use, you are never going to get burnt by research like this, but you are almost certain to be completely boned by implementation errors.
Unfortunately, key management is still a difficult problem. Without secure keys, even the most secure encryption becomes nothing more than an inconvenience.
You seem to be thinking that when I say "implement", I mean, "building an AES core library from scratch". No. I mean properly using a well tested, totally sound AES library in a way that doesn't leave your application totally boned.
I'm happy to say that systems which try to generate random passwords do tend to generate securely random passwords, though. /dev/random has been a huge help there.
I'm just saying that key management is absolutely not a solved problem yet.
And note that SSL download public keys over insecure networks. Verifying them is why certificates were created
You know this.
What I did was come up with the simplest, most reliable key management system there is (static keying with secure keys), and presumed it to be safe.
Then I pointed out that such an application is still likely to be boned by implementation flaws that have nothing to do with how the keys are established.
Can you please fix the images on your blog post about that :)
Wish I could up vote you twice for this.
Part of that was the cost of getting rid of years of cruft, though. Dropping in a new block cipher which operates on 128-bit blocks using 128-, 192-, and 256-bit keys would probably be much cheaper.
There is no need to hurry.
"... the block cipher standard AES is almost as secure as it was 10 years ago in the strongest and most practical model with a single unknown key. The former standard DES has not seen a major improvement since Matsui’s seminal paper in 1993.
In contrast, the area of hash function cryptanalysis is growing quickly, encouraged by the cryptanalysis MD5, of SHA-0 and SHA-1, followed by a practical attack on protocols using MD5, preimage attacks on Tiger and MD5, etc. As differential cryptanalysis, a technique originally developed for ciphers, was carried over to hash function analysis, cryptanalysts are now looking for the opposite: a hash analysis method that would give new results on block ciphers."
Note that you should be using accepted standards for key management anyways. NIST believes that AES-128 has 128 bits of strength, and AES-256 has 256 bits of strength [1, page 63], which they find "Acceptable" until 2031 [1, page 65]; the catch is that any individual key should be expired after a few years. This is called a "cryptoperiod," and the longest they recommend using a key is five years [1, page 55].
By this standard, keep using AES-256, but change your keys every year, and be prepared to migrate to a new cipher at some point within the next decade or two.
[1]: http://csrc.nist.gov/publications/drafts/800-57/Draft_SP800-...