Is a Pronounceable pwgen Password Still Secure in 2026?

What Linux pwgen actually promises

Linux pwgen is a command-line password generator with a human-oriented pronounceable mode and a -s / --secure option for completely random output. Its defaults also depend on whether it is writing to a terminal or a pipe. The manual specifically cautions against using its ordinary, non--s output where an attacker can carry out offline brute-force guessing. See the pwgen 2.08 manual.

That does not mean every output with the name “pwgen” has one fixed security level. Options, length, version and the actual generation algorithm matter. The old pwgen.hu server used that executable; today’s browser-based pronounceable generator uses a different documented construction. It is not an algorithmically compatible port, and its entropy figures must not be applied to arbitrary Linux pwgen output.

Eight characters do not leave much room

A pronounceable construction restricts possible strings so that they have some rhythm. Those restrictions can make typing easier, but they also give an attacker a smaller set to search if the algorithm is known. Counting all letters, digits and symbols that might appear and raising that total to the password length would ignore the restrictions.

For a concrete comparison, our generator’s unsuffixed model has 150 possible two-letter units. An eight-character value uses four units, so it has 150⁴ = 506,250,000 possibilities, about 28.9 bits. The figure is specific to this site’s model, not a claim about the Linux phoneme generator. A 30-character value uses 15 such independent units and reaches about 108.4 bits. Longer output changes the search space substantially because every extra unit introduces another independent selection.

A hypothetical, fully uniform 20-character password from 74 symbols has about 124.2 bits. Our normal password tool additionally enforces selected-group rules, so it reports a conservative lower bound instead of claiming that exact unconstrained figure. Both comparisons are about the generator’s known process, not whether a sample looks complicated.

The attack happens somewhere

Online guessing means sending candidate passwords to a service. Rate limits and other verifier controls constrain how often guesses can be tested. Offline guessing becomes possible when an attacker obtains password-verification data and can test candidates on their own hardware. A suitable salted, costly password hash makes each test more expensive, but the attacker is no longer governed by the website’s login attempt counter.

NIST’s authentication guidance distinguishes these threats, requires rate limiting for password verifiers and addresses protected password storage. It also permits password managers and discourages arbitrary composition rules for user-chosen passwords. These are complementary system requirements, not properties a generator can supply. Read NIST SP 800-63B-4, password verifiers.

Without knowing the storage method, attack budget, available hardware and operational controls, a universal “years to crack” number would be misleading. This site deliberately avoids one. A short value does not become suitable for a valuable account merely because one imagined online scenario would take a long time.

One capital and one digit are not a rescue plan

If a person capitalizes a familiar first letter and appends a favorite digit, the visual complexity increases without introducing much uncertainty. Predictable placement is public structure. Even a genuinely random decimal digit offers only ten choices—about 3.3 bits—when it is appended in a fixed position. A fixed capitalization change adds none.

In our pronounceable tool, uppercase means capitalizing the first letter, and the interface says that it adds no entropy. Optional digit and symbol suffixes are random, but their positions are fixed and their actual choices are counted. They occupy positions within the total selected length, so replacing a letter unit with a suffix can even lower the resulting estimate. A longer base is the meaningful control; a checkbox is not a security certificate.

The random-character generator handles selected-group requirements differently: it draws required characters and cryptographically shuffles all positions. Even there, constraints affect the distribution, which is why its estimate is conservative. The local-generation article explains the selection and shuffle.

Compare the processes, not the marketing labels

These are theoretical output spaces under trusted randomness. They do not measure phishing resistance, device compromise, reuse or a server’s hashing implementation. Capitalization, symbols and word-length preferences can change the model; consult the live estimate for your chosen settings.

Human-invented and machine-generated are different

A machine-generated sequence is sampled from an explicit pool. A person tends to reach for familiar sounds, names, patterns and endings. You cannot assign our model’s entropy to a password you compose yourself just because it follows a similar sound pattern. Repeatedly rejecting outputs until one resembles a favorite name can also change the distribution toward your preferences.

The same distinction applies to passphrases. A six-word quotation is not six independent word selections. The passphrase generator uses the bundled EFF list and allows chance repetitions. You may develop a mnemonic after generation, but keep the actual randomly selected words. EFF’s password guidance explains the role of randomly chosen words and password managers.

Where a shorter code can fit

A short pronounceable value may be tolerable as a temporary, low-risk enrollment code when the receiving system limits attempts, expires it quickly, allows only one use and protects the surrounding process. Those conditions need to be enforced by the system, not assumed by whoever generates the code. A printed workshop access code and a long-lived credential for financial records have different consequences.

Do not use the eight-character preset as a password-manager master password, a disk-encryption secret, an administrator credential or a long-lived API secret. Situations where an attacker can test guesses offline deserve particular care. When a legacy service insists on a short password, use every additional protection it offers and consider whether the service is appropriate for the information involved.

A practical choice for important accounts

Start with a unique password for each account. For most accounts, generate a long random-character password and let a password manager store and fill it. For a secret that you must type or remember, consider a sufficiently long random-word passphrase. If you prefer pronounceable letter sequences, use the longer default and read the displayed estimate rather than falling back to the eight-character habit.

Enable MFA where appropriate, or use a passkey when the service and your recovery arrangements support it. These address parts of the authentication problem that password length alone cannot solve. CISA’s Secure Our World guidance brings together strong passwords, password managers, MFA and other practical protections.

Finally, generating a better string does not change an existing credential automatically. Update it through the real service, save it in your manager, and keep recovery methods current. The pwgen.hu tool gives you local random material and an auditable model. The rest of the account’s security still depends on where and how you use it.