Is it possible to safely check for certain characters in a password?

I'm assuming they're plain text. There's is no perceivable way they can only use those data points to to figure out which hash it is. Unless of course they're using their own "hashing" function which isn't secure at all since it's probably reversible.
- Theoretically they could take those two characters + a salt and then also store that hash. So there it is technically a way to do it although it'd be incredibly redundant, just ask for the actual password at that point.
  
  Please don’t do that. Brute force attacks are very easy on single characters, even two of them.
- Perhaps they validate the passwords client side before hashing. The user could bypass the restrictions pretty easily by modifying the JavaScript of the website, but the password would not be transmitted un-hashed.
  It is worth pointing out that nearly any password restriction like this can be made ineffective by the user anyway. Most people who are asked to put a special character in the password just add a ! to the end. I think length is still a good validation though and it runs into the same issue @randombullet@lemmy.world is asking about
  
  How would they validate individual characters client side? The set password is on the server.
- They could hash the xth letters in a seperate column that are hashed separately, but it's likely they are just storing plain text.

I have never heard of anything secure doing that. Assuming they have taken security steps, it would mean they recorded those characters in plaintext when you set your password, but that means that at least those characters aren't secure, and a breach means some hacker has a great hint.
When the hashing occurs, it happens using the code you downloaded when you visit the site, so it's your computer that does the hash, and then just the hash is sent onwards, so they can't just pull the letters out of a properly secure password.
A secure company would use two-factor authentication to verify you above and beyond your password, anyway, since a compromised password somewhere else automatically compromises questions about your password.
- A lot of banks in the UK do it. They normally have a secondary pin that they will ask for 2 or 3 characters of.
  This means that if you log in and get keylogged/shoulder surfed etc they don't get the full pin. The next time you login you will get asked for different characters.
  Not great, but not awful either - going away now that 2fa is more common
  
  A secondary pin is a bit better but characters from the actual password (that you have to enter anyway) adds nothing to security from that kind of intrusion.
  
  This means that if you log in and get keylogged/shoulder surfed etc they don't get the full pin. The next time you login you will get asked for different characters.
  This seems somehow worse than simply requiring the same few characters each time, since they would either have to store the complete passwords in plaintext, or compute and store the hash for every permutation of 2-3 characters, which is wildly inefficient. You'd also be susceptible to leaking your password if for some reason you are under long term surveillance, since at some point you would presumably have provided all of the characters making up the password.
- Unless they hash and store various combinations of characters in addition to, or instead of, the whole password. I haven't heard of anyone doing this. If you were to pad them with a unique salt and a pepper before hashing each combination, ~~you could end up with something more secure than just hashing the whole password~~ Edit: I was wrong it seems; you'd still end up with something insecure. But hashing the whole password, if done properly, is already secure enough so this would seem like needless complication unless there's some unusual concern about the password being intercepted in transit, and in that case you'd have other problems anyway.
  I have heard of this thing of asking for selected characters of a static second authentication factor (e.g. a PIN), but not of a password itself. And now that we have proper 2FA systems I haven't seen anything like that in a while.
  
  It'll be less secure.
  If they hash a subset, then those extra characters are literally irrelevant, since the hash algorithm will exclude them. Like if they just hashed the first 5 characters, then "passw" is the same as "password" and all those permutations. Hashing is safe because it's one-way, but simple testing on the hashing algorithm would reveal certain characters don't matter.
  Protecting a smaller subset of characters in addition to the whole password is slightly better but still awful. Cracking the smaller subset will be significantly easier using rainbow tables, and literally gives a hint for the whole password, making a rainbow table attack significantly more efficient. Protecting the whole thing (with no easy hints) is way more secure.
  It also adds nothing to keylogging, since it's not even a new code, it's part of the password.
  There was a time where that level of security was acceptable, and it still could be ok on a closed system like an ATM, as the other reply to my comment pointed out, but this kind of protection on a standard computer is outdated and adds holes.

I’ve noticed a few companies ask for specific characters of my password to prove who I am (eg enter the 2nd and 9th character)
They what?!
This is a huge red flag and should not even be possible for your primary password, if they are following basic security principles. Are you sure this isn't a secondary PIN or something like that?
- NatWest in the Uk does it for both the password and the pin, has been since I signed up like 10 years ago. I assumed they do it so you don’t enter a full password that someone could access later. No idea how they work out but they are big and I assume if it was insecure they’d have had issues by now. I assume they store the letter groupings in advance.
  
  I assume if it was insecure they’d have had issues by now.
  At this point, it's okay to assume that they have had issues and they haven't disclosed them.
  
  I would assume they have the whole password in plain text, then. Not much you can do about it, just make sure you're not re-using any part of that password for other services. And if you are, then you should start changing them all to something unique, ideally with a password manager like Keepass or BitWarden. This is a good habit anyway, because you can never really know how companies are handling their IT security.

Do they always ask for the same characters? I'd imagine they could hash the password as well as saving only the 2nd and 9th characters as plaintext. Still a bit of a security risk but not nearly as bad
- Theoretically they could hash the the two characters with a salt and store it that way, but extremely unlikely they'd actually do that. And also fairly pointless. But still technically possible.
  
  I wrote a comment about this in this thread. Just to add that a salt doesn't add complexity to the brute forcing of a password, it just makes it so you need to brute-force each one separately.
  Hashing pairs of characters would be extremely insecure.
- It's always 2 characters, but I can't remember if it is the same ones every time

- Shamir's secret sharing, which was new to me, still means the password must be unencrypted though!? Otherwise there's no secret that can be shared. You can't get individual characters of non-reversible-hashed passwords.
  Reading the Wikipedia page about Shamir's secret sharing I don't see anything about sharing part of the secret data, only that the decryption key is split-shared.
  
  [This comment has been deleted by an automated system]

Good question. A lot of banks in Europe use this type of setup, where it will ask you for 3-4 characters of your pin/password, both to login and to confirm transactions. I always thought it was weird but never thought about the security implications.
- I do have this with my bank as well, but I have always had to enter a full username and different password before it asks for those

One of the main differences between hashing and encrypting is that encryption is réversible by some means, while hashing isn't. The irreversibility is what makes it so ideal for storing a password in a way that definitely can't be used to get the original password back, even if someone steals the whole database with the passwords in it.
Those companies that ask for specific characters might be encrypting the passwords, but they definitely aren't hashing them.

I always figured they checked the plaintext locally before hashing and sending it to their server, but I don't really know.
- This is it.
  It seems a lot of people have trouble distinguishing between what's simply happening in their browser, and what's being sent back to the server. I mean, I get it; it's confusing, even to the people creating the tech, let alone a casual netizen. It's a good question, and you can't fault anyone for wondering what's what.
  
  Yes, asking these questions is a fantastic thing.
  Speaking of questions - I imagine there is a day to use the built in dev tools in the browser to verify that the particular site does this, but I don't know how. Do you happen to know how I might?
  I remember signing up for a site a few years ago and they emailed me my confirmation, with my password, in plaintext. I was absolutely shocked
- that doesn't explain the scenario described in the post
  
  You're right, I misread the post. What sites have done that? I've been fortunate to never encounter any.

They could hash pairs of characters on password creation and store those. Seems like more data points to guess the original password, but maybe the math is hard enough it doesn't do much.
- If it's uniquely salted (and especially if they use a secret pepper too) it might not help with guessing the password, because now you have to crack several hashes of long random sequences of bytes instead of just one.
  
  I'm finding it hard to see how it would be more secure. If I understand what the other comment meant, they would have something like:
  password123 = ef92b778bafe771e89245b89ecbc08a44a4e166c06659911881f383d4473e94f
  We'll assume they pick 4 random pairs
  
  3rd + 5th (sw) = 7865b7e6b9d241d744d330eec3b3a0fe4f9d36af75d96291638504680f805bfd 9th + 11th (13) = 3fdba35f04dc8c462986c992bcf875546257113072a909c162f7e470e581e278 2nd + 5th (aw) = f5fe88ee08735ae259265495a93c8de2b0eacfecf0cd90b5821479be3199fa8c 6th + 8th (od) = 32f30ea0e83c41a331c4330213005568675f7542c25f354ebb634db78cc30d12
  Assuming all 128 7bit character options are used and ignoring dictionary or optimized attacks the complexity of the full password is 7x11 or 77 bits (or 151,115,727,451,828,646,838,272 combinations). So with just the password hash that's how many tries you need to exhaust every possible option, again without optimizing the process.
  But for each of the pairs the complexity is 14 bits or 16,384 combinations. So it would take microseconds to crack all 4 of the pairs. With that information you'd get a password of ?as?wo?d1?3?????? (because we don't know the length) and if they have used a common word or series of words you might have enough information to guess the rest, but even when brute forcing you've removed a decent amount of complexity.
  Note: This is SHA256. We're going to ignore salt for this. Salt only increases complexity because you need to crack each user's password and not able to really use rainbow tables etc.
  Unless I misunderstood the idea. In which case, sorry about that.
  In all likelihood it is encrypted in a database and the interface to the phone operator only allows them to enter what is said and confirm (although I wouldn't be surprised of some showing the whole password).
  
  If you have the password hashes, you almost certainly have the salts. Salts prevent the same password used by different users having the same hash, but if you're bruteforcing, they don't really add to complexity.
  Bruteforcing 2 characters + a salt is computationally the same as bruteforcing 2 characters.

There's a security exchange thread on it here
It looks like there are certain kinds of algorithms you can run that give you this property.
Also, I've seen this when you have an alternate form of authentication like a password you type in full, or an existing session token. In those scenarios, you could probably use some sort of symmetric key encryption to encrypt the secondary password with the primary password / session token in such a way that you aren't storing the key and can't decrypt it, but that you can check specific digits on command.

Full-stack dev here, not necessarily in answer to OP’s question, but in my experience it is a pretty standard practice that when you log in to a service, the web page sends your unhashed creds to the server, where your password is then hashed and compared to the stored hash. Via HTTPS/TLS/SSL, this is a reasonably secure practice since the creds are still encrypted while in transport. Hashing is a computationally expensive process that (before the advent of WASM) wasn’t really feasible to do on the client side.
- What is WASM ?
  
  Web Assembly. Pretty neat tech if you read up on it.

This is typically used for a secondary authentication factor in most instances, such as a service ID number after you've already provided a username, and entered at the same time as a password.
It is not possible for them to securely be storing the password yet being able to retrieve individual characters at the same time.

Something fun you can do is set your password to an eicar test string. That should break things of they are running any av and storing the password in clear text

Any password restriction that LOWERS entropy is a bad restriction.
No, I'm not answering the question, I know. But I'm answering the better question, which is "is this a stupid thing to ask of a password?", and yes it is.

The way I’ve seen this done before, it was not for a password, but for a token entered after the password was entered. No, the token is not encrypted, and therefore it would not be secure for the service to use it as a password.

I’ve only seen this as a second factor after entering a full password. Although it has mostly been replaced by actual 2FA now. Last time I remember this type was on the uk gov student finance website
- The 2 occasions I can think of, it was characters from my main password. Both were during contact with the Support teams. I no longer have service with either of the companies (due to unrelated reasons)
  
  Did you have to give those characters directly to the support staff?

The most secure way this could happen is them storing the specific character separately. It reduces security of your password if that plaintext character is compromised but you could still store the rest of the password securely.
You could even salt and hash the one character with a large salt to keep it behind a one-way function, and then the agent would need to enter it and confirm via the system, but that would reduce any downside of the one or two characters being compromised.
It's weird either way though.

There are at least two components relevant for entering credentials:
the webclient running on your pc/phone
the company server
You open a registration page on a website and your browser downloads the application to your device (just like downloading an app from a store).
The application gives you a form to create an account or a login page where you enter the password/username.
The client then checks if the credentials you entered contain the correct amount of special characters and numbers.
Only after that the credentials are hashed/encrypted and send to the server.
- You didn't read the post's description.
  How is it possible that a company asks you to provide the second letter of your password to prove identify? Surely this would either mean the passwords are stored unencrypted. Or they're using some other form of very dubious security.
  
  Sorry, I misunderstood your post. They could do this by doing the process I described above AND storing parts of your password. Without further research, this sounds insecure though.
  E.g. your password is: SuperSecure? The hash of your password is: 15837A4C3B Your client sends the hash and the characters 0,2,4 to the server: S p r They can then ask you for single digits of the password.
  Another possibility is that they could encrypt the characters with a key to build a more secure value to transmit. As the others have said, this is possibly unsecure too.
  In general: it's services are either comfortable to use or secure. You can achieve both on a basic level, but not on a higher level.