Two very different attacks
When people say “my account was hacked”, one of two things usually happened, and neither involved a hooded figure guessing your password character by character.
Online attacks: guessing through the front door
The attacker submits guesses to the real login form. Every serious service rate-limits this to a few attempts per second per account and locks out after a handful of failures, so the attacker gets perhaps a few thousand tries. That is enough to catch 123456 and Password1, not enough for anything else. Online attacks are why the common password checker exists: if you are in the top few thousand, you fall here.
Offline attacks: the database leaked
The site was breached and its user table copied. If the site stored passwords in plain text, there is nothing to crack. If it stored hashes, the attacker now has a file of fingerprints and can test candidates against them on their own hardware, with no rate limit, at whatever speed the hash function allows. This is where “cracking” actually happens.
The cracking pipeline
Tools such as hashcat and John the Ripper have refined this over twenty years. The stages, in the order they are run because each is cheaper than the next:
- Credential stuffing. Before cracking anything, try every email + password pair from every previous leak against the new site. Billions of pairs are public. This alone breaks most reused passwords, which is why reuse is the single biggest risk.
- Wordlists. Try every password from previous breaches (the “rockyou” list alone has 14 million), every dictionary word in a dozen languages, names, places, sports teams, song lyrics.
- Mangling rules. Apply transformations to every wordlist entry: capitalise the first letter, append 1 through 2029, append
!, swap a→@, e→3, o→0, reverse, duplicate. A rule set with a few thousand rules turns a 14-million-word list into tens of billions of candidates that cover almost everything a human would invent. - Masks. Brute-force with structure: “one capital, five lowercase, two digits” is a mask covering 26 × 26⁵ × 100 ≈ 31 billion candidates, trivial on a GPU. Masks encode the shapes people actually use.
- Pure brute force. Every combination of every character. Only viable for short passwords and fast hashes, but the math is what the crack time calculator shows.
Notice that your clever symbol substitution appears in stage 3, not stage 5. Cracking tools apply it to every word automatically. It costs the attacker one rule.
Speed: the hash function decides
An 8-GPU rig in 2026 tests roughly:
| Hash | Guesses per second | 8 random lowercase chars (209 billion) |
|---|---|---|
| NTLM (Windows) | ~200 billion | about 1 second |
| MD5 | ~100 billion | about 2 seconds |
| SHA-256 | ~10 billion | about 20 seconds |
| bcrypt, cost 12 | ~10 thousand | about 8 months |
| Argon2id, 64 MB | ~1 thousand | about 7 years |
The same password, the same attacker, and a ten-million-fold difference in outcome depending on a decision the website made. This is why how sites store passwords matters as much as the password itself, and why slow hashes exist.
Rainbow tables and why salt killed them
A rainbow table is a precomputed list of hashes for billions of candidate passwords. Against an unsalted hash, cracking becomes a lookup: no computation, instant results for anything in the table. A salt, a random value stored with each hash and mixed into it, means every user’s hash must be attacked separately from scratch, so precomputation buys nothing. Any system built after about 2005 salts; the ones that did not (LinkedIn 2012, 6.5 million unsalted SHA-1 hashes) fell in days.
What this means for you
- Never reuse a password. Stage 1 of every attack is trying what already leaked.
- Length and randomness push you out of stages 2 to 4 and into stage 5, where the numbers are on your side. Sixteen random characters or five random words is enough.
- Assume the fast-hash scenario when you do not know how a site stores passwords, and choose accordingly.
- Turn on two-factor authentication. A cracked password is then not enough on its own.
Check whether any password you use has already fallen with the breach checker, then see how long your current one would survive with the crack time calculator.