Password entropy and passphrases
What a password's entropy measures, and why a long passphrase holds up better than a short word full of symbols.
On this page
In 30 seconds
- Where it happens
- Every time a password is created or changed, at work or at home.
- What the attacker wants
- To recover the password by first trying dictionary words, first names and their classic derivations[1].
- The reflex
- Choose a long passphrase made of randomly drawn words, different for each service[2][3], store it in a password manager and turn on two-factor authentication when it exists[3].
- If it is already done
- If an account is in doubt, or if a service where you have an account has been hacked, change the affected password immediately[3].
What it is
A password's entropy measures its share of randomness. The CNIL defines it as « la quantité de hasard » (the amount of randomness): the theoretical degree of unpredictability of the password, and therefore its ability to withstand a brute-force attack[1].
For an employee, the useful idea fits in a few words: length and randomness matter more than symbols. The UK's NCSC advises combining three random words, for example « applenemobiro », rather than following the old advice of stuffing a word with special characters and digits[2].
Straight answers
What does a password's entropy measure?
It measures the share of randomness. The CNIL recommendation expresses it in bits and sets a generic minimum of 80 bits of entropy for a password with no additional measure[1].
This calculation holds for a randomly generated password. Any construction rule reduces the space of possible choices, and therefore the entropy for a given length, and choosing a password from the words of a language very strongly limits the letter combinations that are possible in practice[1]. NIST writes that « estimating entropy for user-chosen passwords is challenging » and for that reason adopts an approach based mainly on length[4].
Why does a passphrase hold up better than a short word full of symbols?
The CNIL gives three policies that are equivalent in entropy: 12 characters mixing uppercase, lowercase, digits and special characters taken from a list of at least 37, or 14 characters including uppercase, lowercase and digits, with no mandatory special character, or a passphrase of at least 7 words[1].
Symbols add little when they follow habits. Replacing the letter « o » with a zero does not make the password noticeably stronger, because cybercriminals know these tricks, and it makes the password harder to remember[2]. NIST lists length among the main factors in the strength of a password: passwords that are too short give way to brute-force and dictionary attacks[4]. OWASP specifies that long passphrases are effective because they increase the number of attempts that the attacker's dictionary or word list must cover, and not because of a specific entropy value[5].
How is a cracking time calculated?
It depends on two quantities: the number of candidates the attacker has to try, and the rate at which they try them. The rate changes depending on whether the attack is online or offline.
Online, the service can slow the attacker down. The NCSC describes an increasing delay between two attempts and the locking of the account, with 5 to 10 attempts allowed before the freeze[6]. The CNIL draws thresholds from this: a minimum of 50 bits of entropy when access to the account is restricted (delay, maximum number of attempts, captcha, lockout after 10 failures), against 80 bits with no additional measure[1].
Offline, nothing slows the attacker down any more. If a file of stored passwords is stolen, NIST notes that the attackers' ability to compute many billions of hashes per second, with no rate limiting, requires passwords that are several orders of magnitude more complex than those that only have to resist online attacks[4]. The way passwords are stored therefore changes the outcome. OWASP judges fast algorithms such as SHA-256 unsuitable for storing passwords, because they allow many attempts quickly, and recommends slow algorithms, with a unique salt for each password[7]. With graphics cards and servers rented in the cloud, the cost of cracking remains relatively low, especially if these good practices are not followed[7]. Cybermalveillance.gouv.fr, for a brute-force attack, cites « des dizaines de milliers de combinaisons par seconde » (tens of thousands of combinations per second)[3].
The employee knows neither this rate nor how the service stores passwords. What they control is the length and the randomness of their password.
Should you mix uppercase letters, digits and symbols?
Opinions differ. NIST requires services not to impose other composition rules, for example an obligation to mix character types[4]. The NCSC does not recommend complexity rules: they burden users, who fall back on predictable patterns, and they offer no defence against phishing or insecure storage[6]. Cybermalveillance.gouv.fr states that a good password has at least 12 characters mixing uppercase, lowercase, digits and special characters[3].
Should you change your password regularly?
No, unless you suspect a problem. NIST forbids services from requiring a periodic change, but they must force one if there is evidence of compromise[4]. The NCSC considers that regular changing harms security: new passwords are often simple variants of the old ones[6]. The CNIL keeps it only for administration accounts[1]. Cybermalveillance.gouv.fr asks you to change the password immediately at the slightest suspicion[3].
How it works
A password is not stored as it is. NIST requires that it be salted and hashed with a suitable scheme, the purpose of which is to make each attempt more costly for an attacker who has obtained the file[4]. To crack a hash, the attacker picks a password they think is plausible, computes its hash and compares it with the victim's: if they match, the password is known. They repeat the operation with lists drawn from other compromised sites, by brute force or with dictionaries of common passwords[7].
MITRE describes this offline cracking: attackers try to recover usable passwords once they have obtained hashes, using systematic guessing techniques or a precomputed rainbow table. The cracking is generally done on systems the attacker controls, outside the target's network[8]. Against this technique, MITRE cites multifactor authentication and refers to the NIST recommendations for password policy[8].
Dictionary attacks test few candidates, chosen with care: words, first names and their variants. For the word « kangourou », the CNIL gives « k4ng0urou », « kangourou01 » or « KaNgOuRoU » as example derivations[1].
What you do
- Choose a long passphrase made of randomly drawn words. The NCSC starts from three words[2]. The CNIL counts at least 7 words for a passphrase used on its own, with no additional 80-bit measure[1].
- Use a different password for each service, so that a leak makes only the affected service vulnerable[3].
- Use a password manager. The NCSC recommends it: it stores, generates and fills in passwords[6]. Cybermalveillance.gouv.fr cites KeePassXC, which is free, certified by the ANSSI in its version 2.7.9, with a random password generator; only one password is then left to remember, the one that opens the manager[3].
- Turn on multifactor authentication, especially on important accounts: even if the attacker discovers the password, they cannot get into the account without the other factor[6].
- Do not add symbols out of habit: they bring no noticeable strength when they follow known patterns[2].
- Never give a password to a third party. No serious organisation asks for it by messaging or by phone, even for troubleshooting[3].
Pitfalls
- Well-known phrases. A quotation or a proverb is not a random draw: the CNIL's entropy calculation assumes a randomly generated password[1]. The CNIL has added a list of complex but well-known passwords, to be avoided[1].
- Predictable substitutions. Variants such as « k4ng0urou » are among the derivations that dictionary attacks test[1].
- Personal information: dates, favourite sports team, family names or pet names. Most of these details can be found in a social media profile[2]. Logical sequences such as 123456 or azerty are among the first combinations tried[3].
- Reuse. Passwords from leaks are used to attack other systems where the user has used the same one[6].
- The limit of the calculation. OWASP treats entropy estimates as illustrations, not as absolute measures, because they rest on assumptions about the search space[5]. Password strength meters often overlook personal information, repeated characters or common sequences[6]. Finally, keystroke logging, phishing and social engineering succeed just as well against a long, complex password as against a simple one[4].
- Passwords are not the last word. On 23 April 2026, the NCSC published an article titled « NCSC: Leave passwords in the past - passkeys are the future », which presents passkeys as a safer and simpler way for individuals to sign in[9].
Real cases
Night Dragon, February 2011 report
In Night Dragon, the attackers used the Cain & Abel tool to crack password hashes. The case is documented by McAfee in a report dated 10 February 2011[8].
Salt Typhoon, 2025 reports
The Salt Typhoon group cracked passwords protected by weak encryption, obtained from the configuration files of compromised network equipment[8].
Destructive attacks in Poland, 2025
During the « 2025 Poland Wiper Attacks », the attackers tried to crack user passwords[8].
Intrusion into the Mexican government's infrastructure, April 2026 report
In this AI-assisted intrusion, the attacker had Claude Code crack the password hashes from the victim organisations' databases[8].
Sources
- CNIL, “Mots de passe : recommandations pour maîtriser sa sécurité”, 14 October 2022. https://www.cnil.fr/fr/mots-de-passe-recommandations-pour-maitriser-sa-securite (accessed 10 November 2026)
- NCSC (UK), “Three random words”, 21 December 2021. https://www.ncsc.gov.uk/collection/top-tips-for-staying-secure-online/three-random-words (accessed 10 November 2026)
- Cybermalveillance.gouv.fr, “Pourquoi et comment bien gérer ses mots de passe ?”, 23 November 2019. https://www.cybermalveillance.gouv.fr/tous-nos-contenus/bonnes-pratiques/mots-de-passe (accessed 10 November 2026)
- NIST, “NIST Special Publication 800-63B”. https://pages.nist.gov/800-63-4/sp800-63b.html (accessed 10 November 2026)
- OWASP, “Authentication - OWASP Cheat Sheet Series”. https://cheatsheetseries.owasp.org/cheatsheets/Authentication_Cheat_Sheet.html (accessed 10 November 2026)
- NCSC (UK), “Password policy: updating your approach”. https://www.ncsc.gov.uk/collection/passwords/updating-your-approach (accessed 10 November 2026)
- OWASP, “Password Storage - OWASP Cheat Sheet Series”. https://cheatsheetseries.owasp.org/cheatsheets/Password_Storage_Cheat_Sheet.html (accessed 10 November 2026)
- MITRE ATT&CK, “Brute Force: Password Cracking, Sub-technique T1110.002 - Enterprise | MITRE ATT&CK®”. https://attack.mitre.org/techniques/T1110/002/ (accessed 10 November 2026)
- NCSC (UK), “Password administration for system owners”, 19 November 2018. https://www.ncsc.gov.uk/collection/passwords (accessed 10 November 2026)
Cite this article
Brain, “Password entropy and passphrases”, Cyberpedia, 2026. https:/
BibTeX
@misc{brain_cp202_en_2026,
author = {Brain},
title = {Password entropy and passphrases},
year = {2026},
howpublished = {Cyberpedia},
url = {https://startbrain.io/cyberpedia/en/articles/password-entropy-vs-passphrase/},
note = {CC BY 4.0}
}