Brute force attacks are the simplest way into a Windows server: a program guesses usernames and passwords, over and over, until one works. Nobody is sitting at a keyboard. Bots scan the internet for anything that answers on the Remote Desktop port and start guessing as soon as they find it. If your office server is reachable over Remote Desktop from outside, so your staff can open Tally from home during tax season, it is almost certainly receiving these guesses today.
This guide explains how brute force attacks work, what they look like in your server’s logs, and what actually stops them, in the order that matters.
Why this matters for a small office
Remote Desktop is a favourite way in. In the incidents Sophos investigated during 2023, attackers abused Remote Desktop in 90% of cases, and external remote services such as RDP were the first way in for 65% of them (Sophos Active Adversary Report, 2024). Across all breaches, Verizon found stolen or guessed credentials were the most common starting point, at 22% (Verizon DBIR, 2025).
The attacker does not know or care that you run a 12-person CA practice. Your server simply showed up in a scan. If one guess succeeds, they have the same access as the person whose password they guessed: client files, Tally data, email, and often enough rights to switch off backups before deploying ransomware.
How brute force attacks work
“Brute force” covers a family of guessing techniques. The bot picks one depending on what it knows about you, which is usually nothing:
- Dictionary guessing. The bot works through lists of common and leaked passwords. Indian office favourites such as “Tally@123”, “Welcome@123” or “Firm@2026” are on those lists already. If any of your passwords look like these, change them.
- Credential stuffing. Real username and password pairs stolen from other websites are tried against your server. It works when staff reuse a password they also use elsewhere.
- Pure brute force. Every combination in turn. Slow and rarely used against Remote Desktop, but it finishes quickly against short passwords.
- Password spraying. The reverse of the usual approach: one likely password is tried against every account name the bot can think of (administrator, admin, accounts, tally, reception, scanner). Each account sees only one failure, so nothing gets locked.
Bots try the account name “Administrator” early, because it exists on every Windows server. That matters later, when we get to lockout.
Fast floods and slow, low-and-slow campaigns
Brute force attacks come at two speeds, and the difference decides whether simple defences notice them.
Fast floods fire hundreds or thousands of guesses an hour from one address. They are noisy, they trip account lockout, and anyone who opens the log sees them at once. They work only when a password is weak.
Slow, low-and-slow campaigns spread a handful of guesses an hour across many hours or days, often from several addresses, so that no single account ever reaches its lockout limit. Over a month that can still add up to thousands of guesses, and not one of them looks unusual on its own. Password spraying is usually run this way.
Lockout policies stop the first kind. They are blind to the second. Catching slow campaigns means looking at failures per address and per account over long windows, which is a job for monitoring rather than for a Windows setting.
What a brute force attack looks like in your logs
From the outside nothing seems wrong. The server runs normally and staff notice nothing. Open Event Viewer > Windows Logs > Security, though, and the attack is written down line by line. Three event IDs tell most of the story:
- 4625: failed sign-in. One for every wrong guess. Our guide to reading Event ID 4625 explains each field and status code.
- 4740: account locked out. Written when guessing trips your lockout limit.
- 4624: successful sign-in. Logon type 10 is a Remote Desktop session; type 3 is what an RDP sign-in looks like at the network stage when Network Level Authentication is on. A 4624 from an address that has just failed many times is the one you cannot ignore.
An illustrative stretch of a Security log during a night-time attack might read like this:
| Time | Event ID | Account | Source IP | What it means |
|---|---|---|---|---|
| 02:14:03 | 4625 | administrator | 203.0.113.45 | Wrong password |
| 02:14:04 | 4625 | admin | 203.0.113.45 | Account name does not exist |
| 02:14:06 | 4625 | accounts | 203.0.113.45 | Wrong password for a real account |
| 02:31:40 | 4625 | accounts | 198.51.100.17 | Same account, new address |
| 02:31:41 | 4740 | accounts | – | The account is now locked |
| 03:05:12 | 4624 | reception | 198.51.100.17 | A success from an address that was guessing: treat as a break-in |
Each line alone looks harmless. The pattern is the story: many failures, a handful of addresses, account names that really exist in your office, and then a success. To see which addresses are guessing hardest on your server today, run this in PowerShell as Administrator:
Get-WinEvent -FilterHashtable @{LogName='Security'; Id=4625; StartTime=(Get-Date).AddDays(-1)} -ErrorAction SilentlyContinue |
ForEach-Object { ([xml]$_.ToXml()).Event.EventData.Data | Where-Object Name -eq 'IpAddress' } |
Group-Object '#text' | Sort-Object Count -Descending | Select-Object Count, Name -First 10
No output means there were no failed sign-ins in the last 24 hours. If you want to build your own alerting from these events, our brute force detection without a SIEM guide has a complete PowerShell script.
Why account lockout is not enough
An account lockout policy is worth having, but it has three gaps that matter on a small office server:
- The built-in Administrator is exempt. On Windows Server 2022 and older, the built-in Administrator account is not locked out unless the setting “Allow Administrator account lockout” is enabled (Microsoft KB5020282). That is the account bots guess first.
- Slow attacks stay under the limit. Microsoft’s newer default locks an account after 10 failures in 10 minutes. A bot that tries a few times an hour never gets there.
- Lockout can be turned against you. If Remote Desktop is open to the internet, a bot guessing a real account name will lock that person out every morning. Your accounts manager cannot work, and the attacker has lost nothing.
How to stop brute force attacks, in order
- Take Remote Desktop off the internet. This is the step that actually ends the guessing, because a bot cannot guess at a door it cannot reach. Staff can still work from home through Prixl Guard Secure Access (an encrypted tunnel for invited devices only), a VPN or an RD Gateway. If none of those is possible yet, allow Remote Desktop only from known IP addresses. Our guide to restricting RDP access walks through each option, and RDP security explains why it comes first.
- Use long passwords. At least 15 characters, which is the current NIST minimum for a password used on its own (NIST SP 800-63B-4, 2025). Passphrases are fine and easier to remember. Block common and leaked passwords; forced mixes of symbols and capitals add little. More in our password length guide.
- Set an account lockout policy, and include the Administrator. Microsoft’s 10 attempts / 10 minutes / 10 minutes is a sensible baseline. Turn on “Allow Administrator account lockout” too. Remember the third gap above: on an exposed server, lockout can lock out your own staff.
- Give each person their own account. Do not share one “accounts” or “admin” login, and do not use the built-in Administrator day to day. Shared names are easy to guess and impossible to trace.
- Turn on Network Level Authentication. NLA blocks attacks that work before anyone signs in, such as BlueKeep (CVE-2019-0708), and saves server resources. It does not stop password guessing.
- Watch the logs, or have something watch them for you. Prevention reduces the odds. Monitoring tells you when a slow campaign is running or, worse, when a guess has worked.
What about changing the RDP port from 3389? It cuts the number of log entries, because fewer bots find the new port, but it does not stop anyone who looks properly. Treat it as optional noise reduction after the steps above, never as the fix.
What to do if you see this in your logs
Found a wall of 4625 events? Work through this in order:
- Check for successes first. Filter the Security log for Event ID 4624 from the same IP addresses, or against the accounts that were being guessed. Failures are an attempt. A success in the middle of them is an incident: follow our incident response guide, which includes reporting to CERT-In within six hours.
- Block the addresses doing the guessing. Blocking helps for a while, but bots rotate addresses, so it buys time rather than solving anything.
- Close the exposure. The guesses reach you because Remote Desktop is open to the internet. Step 1 of the list above ends them.
- Change the passwords of every real account that was targeted, to something long and unique.
- Look for what an intruder would leave behind: new user accounts (4720), accounts added to the Administrators group (4732) and new services (7045 in the System log). Our post on how to tell if someone is trying to break in lists every sign.
How Prixl Warden helps
Prixl Warden runs as a Windows service on your server and reads the same event logs described above. It tracks failed sign-ins per IP address and per account, so it picks up fast floods and slow, low-and-slow campaigns, and it emails you when it sees one, with a plain-English explanation of what happened. It also flags successful sign-ins outside your working hours, new accounts and changes to the Administrators group.
You can block an attacking address from the dashboard’s IP Blocking tab. Automatic blocking of attacking IPs is opt-in: you switch it on in the dashboard, add your own office and home addresses to Trusted IPs first, and Warden never auto-blocks private (LAN) addresses. On paid plans, alert emails carry a one-click link to block the address. When you are ready to close Remote Desktop to the internet altogether, Prixl Guard gives staff a way in that needs no router changes and no static IP.
Start your 30-day free trial and see what is guessing at your server this week. No card needed.

