It really could just be coincidence, but I've had a handful of users over the last couple days since running 3.1.13 in production swearing up and down they use only one password and they definitely haven't forgotten it. In all the cases I've tested 3.1.13 produces the same results as 3.1.11 that I was running before on my JRuby server, but in #182 @fonica says

updated lib/bcrypt/engine.rb to pass the secret as java bytes; it seems jruby messes up the encoding for certain bytes if the secret is passed as a string.

Could this result in the same password resulting in different hashes in 3.1.12 and 3.1.13? If so, is it just in extreme edge cases or is this something that will happen for sure?