Pull up a chair. You have probably heard someone in a meeting say “we were brute-forced” when what they meant was “someone logged in with a password that was already leaked”. It is a small slip, but it sends the whole team to fix the wrong thing. So let us pull the three attacks apart, slowly, because once you see what each one assumes, the defences almost choose themselves.
Brute force: many guesses against one account
Classic brute force picks one account and tries a very large number of passwords against it. The attacker has no idea what the password is, so every guess is a shot in the dark. If the passwords are drawn from a dictionary of common choices, people call it a dictionary attack; if the attacker walks through every possible combination, it is exhaustive.
Think about what this looks like from the server’s side. One username, hundreds or thousands of failures in a short time. That is loud. Any sensible system notices it, and the standard answer is also simple: lock the account or slow it down after a handful of failures, and add a CAPTCHA or a growing delay between attempts. Against a live login page, brute force is the weakest of the three attacks today.
There is a second place where brute force still matters: offline. If an attacker steals a database of password hashes, no lockout exists because there is no login page. The attacker can guess at the speed of their own hardware. That is a different problem and it is solved by how you store passwords, not by how you protect the login form. Keep those two situations separate in your head.
Password spraying: one guess against many accounts
Spraying flips the shape. Instead of a thousand passwords against one person, the attacker tries one or two very common passwords against thousands of people. Imagine a company with 4,000 employees. The attacker takes a list of usernames (email formats are easy to guess, and staff directories leak) and tries a single seasonal-looking password, the sort that satisfies a rule requiring a capital letter and a digit, against every one of them. Then they wait, perhaps for half an hour, and try the next one.
Why does this work? Because lockout counters are per account. Each account sees one failure, which is nothing. Across 4,000 accounts, though, even if only 0.5 percent use the guessed password, that is 20 accounts handed over. The arithmetic is in the attacker’s favour because they only need a few winners.
The detection idea is therefore different. You stop looking at failures per account and look at failures per source, per time window, across accounts. A single IP address or small set of addresses failing against many distinct usernames is the signature. Attackers know this, so they spread attempts over many addresses, which is why defenders also watch for a rise in overall failure rate, and for successful logins from unusual places right after a burst of failures.
Defences that really help: banned-password lists that block the obvious choices at the moment a user sets a password, multi-factor authentication on everything exposed to the internet (email, VPN, remote desktop gateways), and turning off legacy protocols that cannot do MFA. That last point is where many organisations leak. The main login has MFA, but an old mail protocol endpoint accepts a bare password.
Credential stuffing: the right password, wrong place
Here is the one that is most misunderstood. In credential stuffing, the attacker is not guessing at all. They hold lists of real email and password pairs taken from breaches of other websites. They replay those pairs against your login page, betting that people reuse passwords. Someone who used the same password on a shopping site and on your internal portal has effectively given the attacker a key cut from the old lock.
Notice what is different. Each attempt uses a different, plausible username with its own, plausible password. Most will fail, but the failure rate is far lower than in spraying because the pairs were once valid somewhere. Lockouts do little, because each account sees one or two tries. The traffic is usually automated and spread over a huge pool of addresses, so IP blocking is a game of whack-a-mole.
So what works? MFA again, since a stolen password alone is no longer enough. Checking new passwords against known-breached lists. Rate limits and bot detection at the login endpoint, looking at device and behaviour signals rather than just IP. And education, honestly, though it is the weakest lever: asking people to use a different password for every site only works if you give them a password manager.
A side-by-side you can keep
| Attack | Guesses | Targets | What you notice | Best defences |
|---|---|---|---|---|
| Brute force | Many | One account | Many failures on one username | Lockout or throttling, strong hashing offline |
| Spraying | One or two common ones | Many accounts | Failures across many usernames from few sources | Banned passwords, MFA, retire legacy protocols |
| Stuffing | Known leaked pairs | Many accounts | Spike of automated logins, many sources | MFA, breached-password checks, bot controls |
What your logs should tell you
If you work in a SOC, the habit to build is asking “per what?” Failed logins per account catches brute force. Failed logins per source across accounts catches spraying. A jump in total login volume with a modest success rate, plus successes from fresh devices, points to stuffing. The same raw log, three different questions.
One more thing worth remembering: success is the interesting event. After a burst of failures, look at the first successful login, then at what that account did next. Mailbox rules created, new devices registered, files downloaded in bulk. The attack is not over when the password works; that is where it begins.
So the next time someone says “brute-forced”, ask gently which of the three they mean. It changes the fix, and it changes who needs to act: the identity team, the web team, or the user who reused a password in 2019.
Figures and names in the examples are illustrative.
Leave a Reply