Royce Williams on Nostr: And to your solid point about the 70%, I forgot to math that out to illustrate the ...

And to your solid point about the 70%, I forgot to math that out to illustrate the importance of protecting the remaining people who *aren't* reusing passwords. (And I know you know this, mostly posting for those following along).

In the 16 million bcrypt cost 12 case, if 70% of them can be "pre-cracked" with correlation, then the amount of time for the attack above to run for the remaining 30% drops from 300 days to 90 days. Which is an okay amount of time to buy between compromise and discovery, until we remember that you can have a lot more than two 4090s 😅, and of course, bcrypt parallelizes relatively well, such that a different hash like Argon2 or yescrypt is a better choice. But if bcrypt is the only option for some reason, bumping the work factor higher, if feasible, would be recommended.