How do you tell if someone is trying to break into your server, and whether they have already succeeded? Windows writes down almost everything an intruder does, in the event logs, under numbered event IDs. You do not need to read every entry, only the handful of IDs that separate background noise from an attack in progress, and the one that means the attack worked.
This guide lists those signals in the order an attack usually unfolds, with the event ID for each and a PowerShell command to check your own server in a minute. If you are looking for signs that a server was compromised some time ago, see our companion post on signs your Windows server may have been compromised.
First, separate noise from an attack
If your server accepts Remote Desktop connections from the internet, it receives password guesses from automated bots all day. They come from addresses all over the world, they change constantly, and none of them are aimed at you personally. Chasing each unfamiliar address leads nowhere and wears you out. The useful questions are narrower:
- Are the guesses aimed at account names that really exist in your office?
- Has any address that was guessing ever succeeded?
- Has anything changed on the server that nobody in the office changed?
For an Indian office, “unusual location” is a weak signal anyway. Staff sign in from home in another city, and Jio and Airtel home connections change address often. A new address on its own proves little. A new address that signed in successfully at 2 AM is a different matter.
The signals, with their event IDs
| Signal | Event ID | Log | What it tells you |
|---|---|---|---|
| A wall of failed sign-ins | 4625 | Security | Someone is guessing passwords. Normal background on an exposed server; serious if the account names are real. |
| Staff accounts getting locked out | 4740 | Security | Guesses are hitting a real account hard enough to trip lockout. |
| A success after failures | 4624 (logon type 10 or 3) | Security | An address that was guessing has signed in. Treat as a break-in until proved otherwise. |
| A sign-in at an odd hour | 4624 | Security | A real password used when nobody should be working. |
| A new user account | 4720 | Security | Intruders create their own account so they can come back. |
| Someone added to Administrators | 4732 | Security | A user was added to a local group; check if the group is Administrators or Remote Desktop Users. |
| A new service installed | 7045 | System | Remote-access tools and ransomware often install themselves as services. |
| A new scheduled task | 4698 | Security | Another common way to make malicious software restart on its own. |
| The Security log was cleared | 1102 | Security | Someone is covering their tracks. Very rarely innocent. |
1. Failed sign-ins aimed at real names (4625, 4740)
Every wrong password writes a 4625. On a server with Remote Desktop open to the internet you may see thousands a day. What changes the picture is the account names. Guesses at “administrator”, “admin” or “test” are generic. Guesses at “accounts”, “reception” or a partner’s first name mean the bot has a list that fits your office, or is getting lucky. If a staff member finds their account locked first thing in the morning (Event ID 4740), it is very likely someone guessed at it overnight. Our guide to reading Event ID 4625 explains the status codes that tell a typo from a guess.
2. A success in the middle of failures (4624)
This is the signal that matters most. A 4624 with logon type 10 is a Remote Desktop session; with Network Level Authentication on, an RDP sign-in also shows as type 3 at the network stage. If the source address of a successful sign-in also appears in your failed sign-ins, someone guessed a password. Stop reading and go to the “what to do” section below.
3. A sign-in when nobody should be working (4624)
An intruder with a stolen password often signs in late at night or at the weekend, when nobody will notice. A sign-in at 3 AM is not proof, since your partners may well be working late in filing season, but it is worth one message to the account’s owner: “Was that you?” Our post on after-hours login monitoring covers this in detail.
4. New accounts and new administrators (4720, 4732)
One of the first things an intruder does is create a back door: a new user account, often with an innocent name like “sysbackup” or “support”, added to the Administrators group. If you see a 4720 or 4732 that nobody in the office can explain, treat it as a serious incident. It is also worth listing the members of the Administrators group now and then, so you know what normal looks like:
Get-LocalGroupMember -Group 'Administrators'
5. New services and scheduled tasks (7045, 4698)
Intruders install tools to keep their access: remote-control software, password dumpers, and eventually ransomware. Many register themselves as a Windows service (7045, in the System log) or a scheduled task (4698). A service with a random name, or one running from a user’s Downloads or Temp folder, is a strong warning sign. Remote-support tools such as AnyDesk, UltraViewer or TeamViewer that nobody in the office installed deserve the same suspicion.
6. Security switched off or the log cleared (1102)
Intruders like to work without being watched. Event ID 1102 means someone cleared the Security log, which destroys the record of everything above. A Windows Firewall that has been switched off, or Microsoft Defender’s real-time protection turned off, with nobody in the office admitting to it, belongs in the same category. Check the firewall in a second with:
Get-NetFirewallProfile | Select-Object Name, Enabled
All three profiles (Domain, Private, Public) should show True.
Check your own server in one minute
Open PowerShell as Administrator on the server and paste this. It counts the warning-sign events from the last seven days:
$since = (Get-Date).AddDays(-7)
$checks = @(
@{ Log = 'Security'; Id = 4625, 4740, 4720, 4732, 4698, 1102 },
@{ Log = 'System'; Id = 7045 }
)
foreach ($c in $checks) {
Get-WinEvent -FilterHashtable @{ LogName = $c.Log; Id = $c.Id; StartTime = $since } -ErrorAction SilentlyContinue |
Group-Object Id | Select-Object @{ n = 'Log'; e = { $c.Log } }, @{ n = 'EventId'; e = { $_.Name } }, Count
}
An ID that does not appear in the output had no events in the last week. A large 4625 count tells you the server is exposed. Any 4720, 4732, 7045, 4698 or 1102 deserves a look in Event Viewer to see who and what. New scheduled tasks (4698) are only recorded if “Audit Other Object Access Events” is on; our audit policy baseline covers that.
What to do if you find something
- If an attacker may have got in (a success after failures, an unexplained new administrator, a cleared log): follow our incident response guide. Do not switch the server off, because that destroys evidence. Under the CERT-In Directions of 28 April 2022, a reportable cyber incident must be reported to CERT-In within six hours of noticing it.
- If it is only guessing: close the door. Take Remote Desktop off the internet and give staff another way in, such as Prixl Guard Secure Access, a VPN or an RD Gateway, or at least allow only known IP addresses. Our guide to restricting RDP access walks through each option.
- Either way: change the passwords of every real account that was targeted to something long and unique, and read our guide to stopping brute force attacks.
The problem: nobody watches the logs at 2 AM
Every signal above is visible to anyone who looks. Very few small offices look every day, and intruders count on that. Remote Desktop abuse showed up in 90% of the attacks Sophos’s incident responders handled in 2023 (Sophos Active Adversary Report, 2024), so these are exactly the events worth watching.
Prixl Warden reads these event logs on your server for you: failed and successful sign-ins, brute-force patterns, sign-ins outside working hours, admin sign-ins, new accounts and admin-group changes, lockouts, new services and scheduled tasks, a cleared audit log, and the Windows Firewall being switched off. Serious events are emailed straight away with a plain-English explanation, and a daily digest at 6 PM covers the rest.
Start your 30-day free trial and see what your server’s logs have been saying.

