Okay, so Debian shipped broken OpenSSL for about two years. The advisory is scarce on details (after dealing with crypto literature for a while, you learn to think of “weak RNG” as “well, someone who had perfectly observed twenty billion random numbers I’ve generated could maybe have a fifty-one percent ability to guess the next bit”, which has about zero implications for real world security), but Ben Laurie explains the full story. Apparently, the Debian package maintainer noticed that OpenSSL used uninitialized memory to supply entropy to the RNG, but “fixed” it by disabling the function to add ANY entropy to the pool. This, and the fact that the ssh-vulnkeys utility shipped with the OpenSSH upgrade checks for vulnerable keys by comparing keys to an apparently exhaustive list of some 200000 keys, make it look rather scary. If I understand this correctly, it looks exploitable enough to turn into a nasty real-world problem for some time to come.
On the up side, I really like the engineering of the Debian fix – the new packages refuse to run with compromised keys (and automatically generate safe ones on installation), and upgraded servers refuse logins from unsafe client keys. This eliminates the worst of all the horrors where outdated authorized_keys entries for the unsafe keys would litter the Internets for decades to come, providing rogue access for any idiot with a copy of the master key list…
Update: Ah, the exploit is here.

