ok, after reading that article fully, it does sound a lot less concerning than the headline would like me to believe. it is early in the morning (almost 13:00) and this is a great chance to expose how little i know about all that, so i will:
They believed SSH traffic was immune [...].
classic. we always think that something is perfectly safe until it breaks. also, looking at the article, the issue with RSA has been known since 1996. there had to be a useful application for this. such as TLS. and now some SSH implementations.
Last year, researchers found that [...] they were still able to passively observe faulty signatures that allowed them to compromise the RSA keys of [...] Baidu.com
no idea how this adds any value in a discussion about SSH, but i chuckled.
now the article also get to some more interesting stats.
5.2 billion SSH records. of that 590k with invalid signatures and 4.9k revealed factorization for a total of 189 unique private keys.
now i would very much prefer that last number to be a solid zero, but out of 590k faults, only 4.9k were usable for the attack. everyone that thinks "oh thats nothing. im safe." is still a fool, but it could be far worse. especially since this only target RSA and leaves ed25519 (and others) untouched.
but it just gets even better:
The researchers traced the keys they compromised to devices that used custom, closed-source SSH implementations that didn’t implement the countermeasures found in OpenSSH and other widely used open source code libraries.
if i was drinking something reading this, i would have spat it out laughing. i am that kind of fun at parties. this also partially explains why there are "only" 590k invalid signatures in over 5.2 billion records total. and judging by how good some companies and organizations handle updates (assuming there will be updates from cisco, zyxel, hillstone and mocana), this will still be enough to be used in some attacks five years from now.
