Ask HN: Articles about key rotation being worthless
I need some articles with respect to why the current key rotation recommendations do very little to improve security overall. Given that NIST recommends 1-2 years and others recommend 90-180 day windows, this still gives a disgruntled employee or some other attacker a LOT of time to hack you if they have access to an API key or private key. Does anyone have links to good articles/blogs/white-papers/research about this problem?
7 comments
That's primarily an argument to rotate keys quicker - computers don't care that they have to remember new passwords all the time (which is the main argument against password change requirements: it encourages bad practices from users), so you can do schemes like OAuth2's Refresh Tokens. (and even slow-ish rotation helps with keys forgotten in random places)
I completely agree. But even at 15 or 30 days, it's too long. The only way to protect a key would be to rotate it every day or every hour.
It's steps. E.g. if it's every 15 days, it at least pushes you to the point of automating it (HOPEFULLY) and the app managing it internally - that already helps against stupid shit like "someone put it in code/pushed a config file/... to a repo that later got compromised". Similarly, every X months is still a gain over keys sticking around many years. But yes, at the same time, if you get to have a reliable automated flow there is little reason to not run it with higher frequency.
So, do you know of anyone that has written this type of thing up? I'd love to have some fodder when having these types of discussions. :)
Hm, not specifically. OAuth2 specifications and documentation sort of address the motivation for Refresh Tokens at least (and are widely written about in blog posts etc) - and I think the security recommendations documents now strongly push for Refresh Tokens. For the benefit of automated refresh one could also pull the Let's Encrypt arguments as "similar enough" and widely recognized as good practice.
Just rotate your damn keys. Easy this day with KMS, Vault etc
I agree with you, but it should be WAY faster than every 90 days. I'm trying to find articles that address the fact that NIST and others are worthless since they recommend every 1-2 years.