Research

Analysis of user password strength

The processing power of computers keeps growing, helping users to solve increasingly complex problems faster. A side effect is that passwords that were impossible to guess just a few years ago can be cracked by hackers within mere seconds in 2024. For example, the RTX 4090 GPU is capable of guessing an eight-character password consisting of same-case English letters and digits, or 36 combinable characters, within just 17 seconds.

Our study of resistance to brute-force attacks found that a large percentage of passwords (59%) can be cracked in under one hour.

How passwords are typically stored

To be able to authenticate users, websites need a way to store login-password pairs and use these to verify data entered by the user. In most cases, passwords are stored as hashes, rather than plaintext, so that attackers cannot use them in the event of a leak. To prevent the password from being guessed with the help of rainbow tables, a salt is added before hashing.

Although hashes are inherently irreversible, an attacker with access to a leaked database can try to guess the passwords. They would have an unlimited number of attempts, as the database itself has no protection against brute-forcing whatsoever. Ready-made password-guessing tools, such as hashcat, can be found online.

Methodology

Our study looked at 193 million passwords found freely accessible on various dark web sites. Kaspersky does not collect or store user passwords. More details are available here and here.

We estimated the time it takes to guess a password from a hash using brute force and various advanced algorithms, such as dictionary attacks and/or enumeration of common character combinations. By dictionary we understand here a list of character combinations frequently used in passwords. They include, but are not limited to real English words.

Brute force attacks

The brute-force method is still one of the simplest and most straightforward: the computer tries every possible password option until one works. This is not a one-size-fits-all approach: enumeration ignores dictionary passwords, and it is noticeably worse at guessing longer passwords than shorter ones.

We analyzed the brute-forcing speed as applied to the database under review. For clarity, we have divided the passwords in the sample into patterns according to the types of characters they contain.

  • a: the password contains only lowercase or only uppercase letters.
  • aA: the password contains both lowercase and uppercase letters.
  • 0: the password contains digits.
  • !: the password contains special characters.

The time it takes to crack a password using the brute-force method depends on the length and the number of character types. The results in the table are calculated for the RTX 4090 GPU and the MD5 hashing algorithm with a salt. The speed of enumeration in this configuration is 164 billion hashes per second. The percentages in the table are rounded.

Password pattern Share of passwords of this type in the dataset, % Share of brute-forceable passwords (by pattern, %) Maximum password length in characters by crack time
< 60 s 60 s to 60 min 60 min to 24 h 24 h to 30 d 30 d to 365 d > 365 d 24 h to 30 d 30 d to 365 d > 365 d
aA0! 28 0,2 0,4 5 0 9 85 9 10
a0 26 28 13 15 11 10 24 11 12 13
aA0 24 3 16 11 0 15 55 10 11
a0! 7 2 9 0 14 15 59 9 10 11
0 6 94 4 2 0 0 0
a 6 45 13 10 9 6 17 12 13 14
aA 2 15 22 11 14 0 38 10 11
a! 1 6 9 11 0 11 62 10 11
aA! 0,7 3 2 12 10 0 73 9 10
0! 0,5 10 27 0 18 13 32 10 11 12
! 0,006 50 9 10 5 6 19 11 12 13

The most popular type of passwords (28%) includes lowercase and uppercase letters, special characters and digits. Most of these passwords in the sample under review are difficult to brute-force. About 5% can be guessed within a day, but 85% of this type of passwords take more than a year to work out. The crack time depends on the length: a password of nine characters can be guessed within a year, but one that contains 10 characters, more than a year.

Passwords that are least resistant to brute-force attacks are the ones that consist of only letters, only digits or only special characters. The sample contained 14% of these. Most of them can be cracked within less than a day. Strong letter-only passwords start at 11 characters. There were no strong digit-only passwords in the sample.

Smart brute-force attacks

As mentioned above, brute force is a suboptimal password-guessing algorithm. Passwords often consist of certain character combinations: words, names, dates, sequences (“12345” or “qwerty”). If you make your brute-force algorithm consider this, you can speed up the process:

  • bruteforce_corr is an optimized version of the brute-force method. You can use a large sample to measure the frequency of a certain password pattern. Next, you can allocate to each variety a percentage of computational time that corresponds to its real-life frequency. Thus, if there are three patterns, and the first one is used in 50% of cases, and the second and third in 25%, then per minute our computer will spend 30 seconds enumerating pattern one, and 15 seconds enumerating patterns two and three each.
  • zxcvbn is an advanced algorithm for gauging password strength. The algorithm identifies the pattern the password belongs to, such as “word, three digits” or “special character, dictionary word, digit sequence”. Next, it calculates the number of iterations required for enumerating each element in the pattern. So, if the password contains a dictionary word, finding it will take a number of iterations equal to the size of the dictionary. If a part of the pattern is random, it will have to be brute-forced. You can calculate the total complexity of cracking the password if you know the time it takes to guess each component of the pattern. This method has a limitation: successful enumeration requires specifying a password or assuming a pattern. However, you can find the popularity of patterns by using stolen samples. Then, as with the brute-force option, allocate to the pattern an amount of computational time proportional to its occurrence. We designate this algorithm as “zxcvbn_corr”.
  • unogram is the simplest language algorithm. Rather than requiring a password pattern, it relies on the frequency of each character, calculated from a sample of passwords. The algorithm prioritizes the most popular characters when enumerating. So, to estimate the crack time, it is enough to calculate the probability of the characters appearing in the password.
  • 3gram_seq, ngram_seq are algorithms that calculate the probability of the next character depending on n-1 previous ones. The proposed algorithm starts enumerating one character, and then sequentially adds the next one, while starting with the longest and most frequently occurring n-grams. In the study, we used n-grams ranging from 1 to 10 characters that appear more than 50 times in the password database. The 3gram_seq algorithm is limited to n-grams up to and including three characters long.
  • 3gram_opt_corr, ngram_opt_corr is an optimized version of n-grams. The previous algorithm generated the password from the beginning by adding one character at a time. However, in some cases, enumeration goes faster if you start from the end, from the middle or from several positions simultaneously. *_opt_* algorithms check the varieties described above for a specific password and select the best one. However, in this case, we need a password pattern that allows us to determine where to start generating from. When adjusted for different patterns, these algorithms are generally slower. Still, they can provide a significant advantage for specific passwords.

Also, for each password, we calculated a best value: the best crack time among all the algorithms used. This is a hypothetical ideal case. To implement it, you will need to “guess” an appropriate algorithm or simultaneously run each of the aforementioned algorithms on a GPU of its own.

Below are the results of gauging password strength by running the algorithms on an RTX 4090 GPU for MD5 with a salt.

Crack time Percentage of brute-forceable passwords
ngram_seq 3gram_seq unogram ngram_opt
_corr
3gram_opt
_corr
zxcvbn
_corr
bruteforce
_corr
Best
< 60 s 41% 29% 12% 23% 10% 27% 10% 45%
60 s to 60 min 14% 16% 12% 15% 12% 15% 10% 14%
60 min to 24 h 9% 11% 12% 11% 12% 9% 6% 8%
24 h to 30 d 7% 9% 11% 10% 11% 9% 9% 6%
30 d to 365 d 4% 5% 7% 6% 8% 6% 10% 4%
> 365 d 25% 30% 47% 35% 47% 35% 54% 23%

The bottom line is, when using the most efficient algorithm, 45% of passwords in the sample under review can be guessed within one minute, 59% within one hour, and 73% within a month. Only 23% of passwords take more than one year to crack.

Importantly, guessing all the passwords in the database will take almost as much time as guessing one of them. During the attack, the hacker checks the database for the hash obtained in the current iteration. If the hash is in the database, the password is marked as cracked, and the algorithm moves on to working on the others.

The use of dictionary words reduces password strength

To find which password patterns are most resistant to hacking, we calculated the best value for an expanded set of criteria. For this purpose, we created a dictionary of frequently used combinations of four or more characters, and added these to the password pattern list.

  • dict: the password contains one or more dictionary words.
  • dict_only: the password contains only dictionary words.
Password pattern Share of passwords, % Share of passwords that can be cracked with a dictionary attack (by pattern, %) Maximum password length in characters by crack time
< 60 s 60 s to 60 min 60 min to 24 h 24 h to 30 d 30 d to 365 d > 365 d 24 h to 30 d 30 d to 365 d > 365 d
dict_a0 17 63 15 8 5 3 7 10 11 12
aA0! 14 5 6 5 5 3 76 6 7 8
dict_aA0 14 51 17 10 7 4 11 9 10 11
dict_aA0! 14 34 18 12 10 6 20 7 8 8
a0 10 59 22 6 6 1.8 6 10 11 12
aA0 10 19 13 13 6 7 42 9 10 11
0 6 92 5 1.5 1.3 0 0 15
dict_a0! 5 44 16 10 8 5 17 9 9 10
dict_a 4 69 12 6 4 2 6 11 12 13
a0! 2 31 19 13 9 5 23 9 9 10
a 1.2 76 7 6 3 3 6 11 12 13
dict_aA 1.2 56 15 8 6 3 11 9 10 10
dict_a! 0.8 38 16 10 8 5 23 8 9 10
aA 0.7 26 10 28 7 2 27 9 10 10
dict_aA! 0.5 31 17 11 10 6 26 8 9 9
0! 0.4 53 15 8 7 5 13 9 10 11
dict_only 0.2 99.99 0.01 0.0002 0.0002 0 0 18
dict_0 0.2 89 6 2 2 0 0 15
aA! 0.2 11 8 10 16 3 52 8 9 9
a! 0.1 35 16 10 9 5 25 8 9 10
dict_0! 0.06 52 13 7 6 4 17 9 10 11
! 0.006 50 10 6 8 4 20 8 9 10

The majority (57%) of the passwords reviewed contained a dictionary word, which significantly reduced their strength. Half of these can be cracked in less than a minute, and 67% within one hour. Only 12% of dictionary passwords are strong enough and take more than a year to guess. Even when using all recommended character types (uppercase and lowercase letters, digits and special characters), only 20% of these passwords proved resistant to brute-forcing.

It is possible to distinguish several groups among the most popular dictionary sequences found in passwords.

  • Names: “ahmed”, “nguyen”, “kumar”, “kevin”, “daniel”;
  • Popular words: “forever”, “love”, “google”, “hacker”, “gamer”;
  • Standard passwords: “password”, “qwerty12345”, “admin”, “12345”, “team”.

Non-dictionary passwords comprised 43% of the sample. Some were weak, such as those consisting of same-case letters and digits (10%) or digits only (6%). However, adding all recommended character types (the aA0! pattern) makes 76% of these passwords strong enough.

Takeaways

Modern GPUs are capable of cracking user passwords at a tremendous speed. The simplest brute-force algorithm can crack any password up to eight characters long within less than a day. Smart hacking algorithms can quickly guess even long passwords. These use dictionaries, consider character substitution (“e” to “3”, “1” to “!” or “a” to “@”) and popular combinations (“qwerty”, “12345”, “asdfg”).

This study lets us draw the following conclusions about password strength:

  • Many user passwords are not strong enough: 59% can be guessed within one hour.
  • Using meaningful words, names and standard character combinations significantly reduces the time it takes to guess the password.
  • The least secure password is one that consists entirely of digits or words.

To protect your accounts from hacking:

  • Remember that the best password is a random, computer-generated one. Many password managers are capable of generating passwords.
  • Use mnemonic, rather than meaningful, phrases.
  • Check your password for resistance to hacking. You can do this with the help of Password Checker, Kaspersky Password Manager or the zxcvbn
  • Make sure your passwords are not contained in any leaked databases by going to haveibeenpwned. Use security solutions that alert users about password leaks.
  • Avoid using the same password for multiple websites. If your passwords are unique, cracking one of them would cause less damage.

Analysis of user password strength

Your email address will not be published. Required fields are marked *

 

  1. john fuller

    Password!

  2. Jeff

    There is an assumption here that the service you’re accessing is using MD5 to hash passwords, which if it is at all serious (e.g. a bank) it absolutely shouldn’t be doing. Any vaguely respectable system would not have been using MD5 in the last decade – they should be using something like Bcrypt which creates much more expensive hashes.

    Still the best advice is there at the end – use a password manager and have different passwords for each site – especially the important ones like your bank.

    1. Alexey Antonov

      This is a essential point. Bcrypt would be much more difficult to crack.
      Unfortunately, in the real world, as we see, many services still use weak hashes. Moreover, in 2024 we still sometimes find password databases in clear text.
      So in this research we’ve used md5 as a baseline.

  3. James Sterling

    Brute-forcing a hash is not “password guessing”.

    1. Alexey Antonov

      Hi James!

      Apart from being a term for a specific MITRE subtechnique, “password guessing” may be used as a general definition of password cracking/brute-forcing, as it is used by Wikipedia: https://en.wikipedia.org/wiki/Password_cracking
      It is also worth noting that MITRE’s page for “password cracking” features the wording “guess the passwords” as well:

      “Techniques to systematically guess the passwords used to compute hashes are available, or the adversary may use a pre-computed rainbow table to crack hashes.”
      https://attack.mitre.org/techniques/T1110/002/

Reports

BlindEagle flying high in Latin America

Kaspersky shares insights into the activity and TTPs of the BlindEagle APT, which targets organizations and individuals in Colombia, Ecuador, Chile, Panama and other Latin American countries.

APT trends report Q2 2024

The report features the most significant developments relating to APT groups in Q2 2024, including the new backdoor in Linux utility XZ, a new RAT called SalmonQT, and hacktivist activity.

Subscribe to our weekly e-mails

The hottest research right in your inbox