Secure passwords and password managers, without folklore
Mandatory special characters and a change every three months are officially history. What actually counts today, and why a password manager is right despite its flaws.
The good news first: most password rules you have wrestled with for years no longer apply. No character soup, no change every three months.
The old rules have been abolished, from the top down
The US standards body NIST published its guideline SP 800-63B-4 in July 2025. It says, word for word: "Verifiers and CSPs SHALL NOT impose other composition rules (e.g., requiring mixtures of different character types) for passwords." So a service may no longer demand an uppercase letter, a digit and a special character. The sentence on forced rotation is just as clear: "Verifiers and CSPs SHALL NOT require subscribers to change passwords periodically." You change it only when there is evidence of compromise.
What counts instead is length: at least 15 characters for a password that protects an account on its own, eight are allowed if it is only one of two factors. And every new password should be checked against a list of known, breached or commonly used ones.
Germany's BSI takes the same view. On 31 January 2025 it wrote that regular, occasion-independent password changes tend to lead to increasingly weaker passwords, and that third parties, employers in particular, should neither demand nor technically enforce them.
Three random words, six is better
So what instead? The NCSC in Britain recommends a method you can actually remember: "Combine three random words to create a password that's 'long enough and strong enough'." On swapping characters, a zero for an O, it is blunt: "cyber criminals know these tricks as well. So your password won't be significantly stronger, but it will be harder for you to remember." The reason for the whole reversal sits in the same text: "memorising lots of complex passwords is almost impossible."
If you want to be precise, use dice. The EFF publishes word lists where five dice pick each word. Its recommendation: "We recommend selecting a minimum of six words from our long wordlist." Six such words are, by the EFF's own count, one of about 2⁷⁷ possibilities. What matters is that the randomness comes from the dice, not from your head: whatever you think up, others think up too.
| Old rule | What applies today |
|---|---|
| upper, lower, digit, special character | no character-class requirement (NIST) |
| change every three months | change only when there is a reason (NIST, BSI) |
| as cryptic as possible | as long as possible, at least 15 characters without a second factor (NIST) |
| never write it down | writing it down is fine if you keep it somewhere safe (NCSC) |
The password manager, warts and all
Nobody can memorise a long, unique password for every account. Hence the manager, which then becomes the single point everything hangs on. Together with the FZI research centre, the BSI examined ten password managers and published on 9 December 2025 that three of them stored passwords in a way that would theoretically let the vendor access them.
LastPass showed that this need not stay theoretical. In December 2022 the company disclosed that an attacker had copied a backup of customer vault data: website addresses unencrypted, usernames and passwords encrypted and openable only with the key derived from the master password, which LastPass says it never knows.
And still the BSI sticks to its recommendation: the flaws are no reason to do without a password manager, and in its view the benefit far outweighs them. The reason is the alternative, because without a manager people reuse passwords and pick weak ones, so the risks of not using one are considerably greater than the implementation flaws of individual products. NIST agrees: verifiers "SHALL allow the use of password managers and autofill functionality", because managers have been shown to lead to stronger passwords.
The one password you have to know yourself
That leaves the master password, exactly what the dice passphrase is made for. The EFF writes that a passphrase is especially suitable "when directly used to encrypt information" and names unlocking a password safe as a case in point. The difference from an online account matters: there, the number of guesses is limited. If an encrypted file sits on the attacker's machine, they keep guessing offline, as fast as their hardware allows.
The BSI spells out the price: lose the master password and, in the worst case, all the data goes with it. So write it down, but keep it where your passport lives. The NCSC agrees: "If you want to write your password down, that's also OK, provided you keep it somewhere safe."
And has yours already leaked?
The Pwned Passwords service uses a method called k-anonymity "to let you check whether a password has been seen before without ever sending the full password or its complete hash to the service". Your browser hashes locally and sends only the first five characters; the comparison happens on your side. That covers only part of NIST's blocklist requirement, which also counts dictionary words and terms around the service itself.
Where we at PriChat get it wrong ourselves
At registration PriChat requires at least eight characters plus an uppercase and a lowercase letter, a digit and a special character. Those are exactly the composition rules NIST prohibits, and eight characters sit well below the fifteen it asks for. We do not check against breach lists at all. Both are in the code as described, and both are open items, not recommendations.
What we do consider right: your password is not a mere nameplate here. It produces the key that encrypts your messages in the browser, via PBKDF2-SHA256 with 600,000 rounds (200,000 for older accounts). That is what makes the offline guessing above expensive. The server never sees your password, only a login token derived from it, which it additionally hashes with Argon2. A second factor via authenticator app and passkey sign-in are available. And we cannot reset your password: only your recovery phrase can, because the server holds only an encrypted recovery block it cannot open.
Length beats character soup, uniqueness beats length. A random long password per account from the manager, a dice-rolled master password in your head, a second factor on top.
What this means for you
- Stop changing passwords without a reason. Change one immediately if a service reports a breach.
- Use a password manager and let it generate the passwords. Which one is secondary; whether you use one at all is not.
- For the master password, and for anything that directly encrypts something: six random words, rolled rather than invented.
- Turn on a second factor, or use passkeys where they exist. Even a good password is only one factor.
Sources
- NIST · SP 800-63B-4, Password Verifiers (Juli 2025)
- BSI · Passwortwechseln war gestern (31.01.2025)
- BSI · Verbesserungsbedarf bei Passwortmanagern (09.12.2025)
- BSI · Passwörter verwalten mit einem Passwort-Manager
- NCSC UK · Three random words (21.12.2021)
- EFF · Dice-Generated Passphrases
- Have I Been Pwned · Pwned Passwords
- LastPass · Notice of Recent Security Incident (22.12.2022)