Here is the text of the NIST sp800-63b Digital Identity Guidelines.
Reworded rules for clarity:
- Min required length must be 8 chars (obligatory), but it should be 15 chars (recommended).
- Max length should allow at least 64 chars.
- You should accept all ASCII plus space.
- You should accept Unicode; if doing so, you must count each code as one char.
- Don’t demand composition rules (e.g. “u’re password requires a comma! lol lmao haha” tier idiocy)
- Don’t bug users to change passwords periodically. Only do it if there’s evidence of compromise.
- Don’t store password hints that others can guess.
- Don’t prompt the user to use knowledge-based authentication.
- Don’t truncate passwords for verification.
I was expecting idiotic rules screaming “bureaucratic muppets don’t know what they’re legislating on”, but instead what I’m seeing is surprisingly sane and sensible.
gosh who would want an uncommon character that obviously most average people aren’t thinking about in their passwords, that sounds like it might even be somewhat secure.
Very common for pass phrases, and not dissuaded. Pass phrases are good for people to remember without using poor storage practices (post it notes, txt file, etc) and are strong enough to keep secure against brute force attacks or just guessing based off knowledge of the user.
On one hand, that’s true. On the other hand, a person should only need exactly one passphrase, which is the one used to unlock their password manager. Every other password should be randomly-generated and would only contain space characters by chance.
I’m with you, despite seeing lemmings downvote the heck out of your comment 😢
The reason, and specifically for whitespace at the beginning or end of a password, is that a lot of users copy-paste their passwords into the form, and for various reasons, whitespace can get pasted in, causing an invalid match. No bueno.
Source: I’m a web developer who has seen this enough times that we had to implement a whitespace-trim validation for both setting & entering passwords.
NIST generally knows what they’re doing. Want to overwrite a hard drive securely? NIST 800-88 has you covered. Need a competition for a new block cipher? NIST ran that and AES came out of it. Same for a new hash with SHA3.
It’s crazy that they didn’t include all the “should” items in that list. If you read the entire section, there’s a critical element that’s missing in the list, which is that new passwords should be checked against blocklists. Otherwise, if you combine 1, 5, and 6, you end up with people using “password” as their password, and keeping that forever. Really, really poor organization on their part. I’m already fighting this at work.
I was expecting idiotic rules screaming “bureaucratic muppets don’t know what they’re legislating on”, but instead what I’m seeing is surprisingly sane and sensible
NIST knows what they’re doing. It’s getting organizations to adapt that’s hard. NIST has recommended against expiring passwords for like a decade already, for example, yet pretty much every IT dept still has passwords expiring at least once a year.
re #7, I hope they are also saying no ‘secret questions’ to reset the password?
Yeah, I think 7 and 8 both cover that. I recently signed up for an account where all of the “security questions” provided asked about things that could be either looked up or reasonably guessed based on looked up information.
We live in a tech world designed for the technically illiterate.
Sarah Palin had her Yahoo mail account hacked because of those “security” questions. In 2008. We should be well past the time where they are a thing.
I usually invent answers to those and store those answers in a password manager. Essentially turns them into backup passwords that can be spoken over the phone if necessary.
Where was I born? “Stallheim, EUSA, Mars”
Name of first pet? “Groovy Tuesday”
It’s fun, usually.
I think it’s pretty idiotic to
Verifiers and CSPs SHALL NOT impose other composition rules (e.g., requiring mixtures of different character types) for passwords.
They might mean well, but the reason we require a special character and number is to ensure the amount of possible characters are increased.
If a website doesn’t enforce it, people are just going to do a password like password
password is a totally valid password under this rule. Any 8 letter word is valid. hopsital for example.
These passwords can be cracked in seconds under 10 minutes, and have their hashes checked for in leaks in no time if the salt is also exposed in the hack.
Edit: Below
Numbers from a calculator with 8 characters using sha2 (ignoring that crackers will try obvious fill ins like 0 for o and words before random characters, this is just for example)
hospital 5m 23s
Hospital 10m 47s
Hospita! 39m 12s
Moving beyond 8
Hospita!r - 19h 49m
Hospita!ro 3w 4d
Hospita!roo 2y 1m
Hospita!room 66 years
The suggestion of multiple random words makes not needing the characters but you have to enforce a longer limit then, not 8.
At least with 11 characters with upper case and special characters if it was all random you get about 2 years after a breach to do something instead of mere weeks. If it was 11 characters all lower case nothing special you’d only get 2 months and we are rarely notified that fast.
They might mean well, but the reason we require a special character and number is to ensure the amount of possible characters are increased.
The problem with this sort of requirement is that most people will solve it the laziest way. In this case, “ah, I can’t use «hospital»? Mkay, «Hospital1» it is! Yay it’s accepted!”. And then there’s zero additional entropy - because the first char still has 26 states, and the additional char has one state.
Someone could of course “solve” this by inserting even further rules, like “you must have at least one number and one capital letter inside the password”, but then you get users annotating the password in a .txt file because it’s too hard to remember where they capitalised it or did their 1337.
Instead just skip all those silly rules. If offline attacks are such a concern, increase the min pass length. Using both lengths provided by the guidelines:
- 8 chars, mixing:minuscules, capitals, digits, and any 20 special chars of your choice, for a total of 82 states per char. 82⁸ = 2*10¹⁵ states per password.
- 15 chars, using only minuscules, for a total of 26 states per char. Number of states: 26¹⁵ = 1.7*10²¹ states per password.
But they mess that up with their 8 char rule
Verifiers and CSPs SHALL require passwords to be a minimum of eight characters in length and SHOULD require passwords to be a minimum of 15 characters in length.
I’d they’d just said shall require 15 but not require special chars then that’s okay, but they didn’t.
Then you end up with the typical shitty manager who sees this, and says they recommend 8 and no special chars, and that’s what it becomes.
Don’t bug users to change passwords periodically. Only do it if there’s evidence of compromise.
This is a big one. Especially in corporate environments where most of the users are, shall we say, not tech savvy. Forcing people to comply with byzantine incomprehensible password composition rules plus incessantly insisting that they change their password every 7/14/30 days to a new inscrutable string that looks like somebody sneezed in punctuation marks accomplishes nothing other than enticing everyone to just write their password down on a Post-It and stick it to their monitor or under their keyboard.
Remember: Users do not care about passwords. From the perspective of anyone who isn’t a programmer or a security expert, passwords are just yet another exasperating roadblock some nerd keeps putting in front of them that is preventing them from doing whatever it is they were actually trying to do.
Everyone I’ve spoken to who has a password change rule just changes one character from their previous password. It does NOTHING.
That works great until some dickhole implements the old, “New password cannot contain any sequence from your previous (5) passwords.”
This also of course necessitates storing (multiple successive!) passwords in plain text or with a reversible cipher, which is another stupid move. You’d think we’d have gotten all of this out of our collective system as a society by now, and yet I still see it all the time.
All of these schemes are just security theater, and actively make the system in question less secure while accomplishing nothing other than berating and frustrating its users.
- Don’t truncate passwords for verification.
It needed to be said. Because some password system architects have been just that stupid.
Edit: Fear of other’s stupidity is the mind killer. I will face my fear. My fear will wash over me, and when it has passed, only I will remain. Or I’ll be dead in a car accident caused by an AI driver.
I’ve seen sites truncate when setting, but not on checking. So you set a password on a site with no stated limit, go to use said password, and get locked out. It’s infuriating
Another ridiculous policy I’ve seen (many years ago) is logging in too fast. I used to get locked out of my banks website all the time and I used autotype with KeePass so I was baffled when it wouldn’t get accepted. Eventually I had a thought to slow down the typing mechanism and suddenly I didn’t get locked out anymore.
Years back, I had that happen on PayPal of all websites. Their account creation and reset pages silently and automatically truncated my password to 16 chars or something before hashing, but the actual login page didn’t, so the password didn’t work at all unless I backspaced it to the character limit. I forgot how I even found that out but it was a very frustrating few hours.
You should accept Unicode; if doing so, you must count each code as one char.
Hmm. I wonder about this one. Different ways to encode the same character. Different ways to calculate the length. No obvious max byte size.
Who cares? It’s going to be hashed anyway. If the same user can generate the same input, it will result in the same hash. If another user can’t generate the same input, well, that’s really rather the point. And I can’t think of a single backend, language, or framework that doesn’t treat a single Unicode character as one character. Byte length of the character is irrelevant as long as you’re not doing something ridiculous like intentionally parsing your input in binary and blithely assuming that every character must be 8 bits in length.
If the same user can generate the same input, it will result in the same hash.
Yes, if. I don’t know if you can guarantee that. It’s all fun and games as long as you’re doing English. In other languages, you get characters that can be encoded in more than 1 way. User at home has a localized keyboard with a dedicated key for such a character. User travels across the border and has a different language keyboard and uses a different way to create the character. Euro problems.
https://en.wikipedia.org/wiki/Unicode_equivalence
Byte length of the character is irrelevant as long as you’re not doing something ridiculous like intentionally parsing your input in binary and blithely assuming that every character must be 8 bits in length.
There is always some son-of-a-bitch who doesn’t get the word.
- John F. Kennedy
It matters for bcrypt/scrypt. They have a 72 byte limit. Not characters, bytes.
That said, I also think it doesn’t matter much. Reasonable length passphrases that could be covered by the old Latin-1 charset can easily fit in that. If you’re talking about KJC languages, then each character is actually a whole word, and you’re packing a lot of entropy into one character. 72 bytes is already beyond what’s needed for security; it’s diminishing returns at that point.
I think if you do allow 8 character passwords the only stipulation is that you check it against known compromised password lists. Again, pretty reasonable.
That stipulation goes rather close to #5, even not being a composition rule. EDIT: see below.
I think that a better approach is to follow the recommended min length (15 chars), unless there are good reasons to lower it and you’re reasonably sure that your delay between failed password attempts works flawlessly.
EDIT: as I was re-reading the original, I found the relevant excerpt:
If the CSP [credential service provider] disallows a chosen password because it is on a blocklist of commonly used, expected, or compromised values (see Sec. 3.1.1.2), the subscriber SHALL be required to choose a different password. Other complexity requirements for passwords SHALL NOT be imposed. A rationale for this is presented in Appendix A, Strength of Passwords.
So they are requiring CSPs to do what you said, and check it against a list of compromised passwords. However they aren’t associating it with password length; on that, the Appendix 2 basically says that min length depends on the threat model being addressed; as in, if it’s just some muppet trying passwords online versus trying it offline.
I hate that anyone has to be told not to truncate passwords. Like even if you haven’t had any training at all, you’d have to be advanced stupid to even come up with that idea in the first place.
Can you elaborate further? Why would someone want to truncate passwords to begin with?
To save a few megabytes of text in a database somewhere. Likely the same database that gets hacked.
Microsoft used to do that. I made a password in the late 90’s for a we service and I found out that it truncated my password when they made it after it warned my my password was too long when I tried to log in. It truncated at 16 characters.
The weirdest one I found was a site that would only check to see if what you entered started with the correct password. So if your password was hunter2 and you tried hunter246, it would let you in.
Which means not only were they storing the password, but they had to go out of their way to use the wrong kind of string comparison.
The LM password hash (predecessor to NTLM) was calculated in two blocks of 7 characters from that truncated 14 characters. Which meant the rainbow table for that is much smaller than necessary and if your password is not 14 characters, then technically part of the hash is much easier to brute force, because the other missing characters are just padded with null.
