Most small offices run one Windows server that matters: the Tally data, the client folders, perhaps a document system, reached over Remote Desktop by staff in the office and, in tax season, from home. You can secure your Windows server in an afternoon. The five steps below close the doors attackers actually use, and each one ends with a quick check so you know it worked rather than hoping it did.
Step 1: Take Remote Desktop off the internet
If Remote Desktop (RDP) can be reached from the internet, password-guessing bots are already working on it. In the incidents Sophos’s responders handled in 2023, external remote services such as RDP were the way in for 65% of cases (Sophos Active Adversary Report, 2024). A strong password only slows attackers down; not being reachable stops them.
Action, in this order of preference:
- Put staff behind a private tunnel. Prixl Guard Secure Access gives each invited person an encrypted WireGuard tunnel to the office server; once everyone is on it, switch on “Lock RDP to Secure Access only” and Remote Desktop is closed to the internet. Its default connection method needs no router setup and no static IP, which suits Jio and Airtel connections. The staff app is Windows 64-bit only today. A VPN or an RD Gateway with multi-factor sign-in does the same job if you already have one.
- If you cannot do that yet, allow-list known addresses. Restrict RDP to the fixed IPs of the office and the people who work from home, in Windows Firewall and on the router or cloud security group.
- Turn on Network Level Authentication (NLA). It blocks pre-authentication exploits such as BlueKeep (CVE-2019-0708) and saves server resources. It does not stop password guessing, so it is an addition to the first two points, not a replacement.
Moving RDP to a different port only reduces the noise in your logs; bots find the new port. If you do it, do it after the real fix, and read how to change the RDP port without locking yourself out first.
While you are on the router’s port-forwarding page, look at everything else that is forwarded to the server. File sharing (port 445), remote management (ports 5985 and 5986) and SQL Server (1433) should never be forwarded from the internet. If the server hosts a website, 80 and 443 may need to be open; almost nothing else does. Delete any forward nobody can explain.
Verify: first find the port RDP really listens on (it may have been moved):
(Get-ItemProperty -Path 'HKLM:\SYSTEM\CurrentControlSet\Control\Terminal Server\WinStations\RDP-Tcp' -Name PortNumber).PortNumber
Then test from outside the office, for example from a laptop on a phone hotspot. Testing from inside the office against your own public IP gives misleading results on many routers. Replace the address and port with yours:
Test-NetConnection 203.0.113.10 -Port 3389
TcpTestSucceeded : False is what you want. If it says True, so can every bot on the internet.
Step 2: Set account policies that hold up
Password-guessing lists are built from millions of real leaked passwords, so anything shaped like “Firm@2026” or “Tally@123” is already on them, symbols and all. Length does more than forced character mixes. NIST’s current guidance sets a 15-character minimum for passwords used on their own, drops the rule that forces symbols and capitals, and asks you to block known-breached and common passwords instead (NIST SP 800-63B-4, 2025). A passphrase of four or five unrelated words meets that and is easier to type.
- Password length. Open
secpol.mscand go to Account Policies > Password Policy > Minimum password length. Windows caps this setting at 14 unless “Relax minimum password length limits” is enabled in the same folder, which needs Windows Server 2022 or newer (Microsoft). On Server 2016 or 2019, set 14 and make 15 the rule in your written policy. Our guide to minimum password length has the details. Warden’s audit flags anything under 10; we recommend 15. - Account lockout. Under Account Policies > Account Lockout Policy, set a threshold. Microsoft’s newer default is 10 failed attempts in 10 minutes, locked for 10 minutes (KB5020282). Two catches. On Server 2022 and older, the built-in Administrator is never locked out unless you also enable “Allow Administrator account lockout” in the same folder. And on a server that is still open to the internet, lockout lets an attacker lock your staff out on purpose, which is one more reason to finish Step 1. More in our account lockout guide.
- The built-in Administrator. It is the first username attackers try. Rename it or disable it, and give each person who needs admin rights their own named account so you can see who did what.
Verify: net accounts shows the minimum password length and lockout threshold. This shows the current name and state of the built-in Administrator (its security ID always ends in -500):
Get-LocalUser | Where-Object { $_.SID -like '*-500' } | Select-Object Name, Enabled
Step 3: Keep Windows and the software on it patched
Attackers reuse old, published holes because so many servers never get the fix. WannaCry spread in May 2017 through an SMB flaw that Microsoft had patched two months earlier in MS17-010. Servers that had installed that update were protected from that route.
Action: turn on automatic updates, or fix one patching day a month (the second Tuesday is Microsoft’s release day) and keep it. Include the software on the server: Tally, the backup agent, any browser or PDF reader someone uses there. Check the Windows version too. Windows Server 2012 and 2012 R2 went out of support in October 2023, and Windows Server 2016 support ends on 12 January 2027. If your server runs either, plan the upgrade now; our post on updates explains why waiting costs more.
Verify: Settings > Windows Update should say you are up to date with a recent check date. This lists the last few updates installed:
Get-HotFix | Sort-Object InstalledOn -Descending | Select-Object -First 5 HotFixID, InstalledOn
Step 4: Cut admin access down to the people who need it
Every admin account is a way in, and every extra one makes a single phished password more expensive. On a small-office server that has run for a few years, the Administrators group often holds surprises: an articled assistant who left two years ago, the vendor who installed the backup software, a generic “admin” nobody remembers creating.
Action: open Computer Management > Local Users and Groups > Groups > Administrators and question every entry. Remove admin rights from anyone who does not need them for daily work, and have admins use a standard account for email and browsing. Vendor access should end when the job does. Disable a leaver’s account on their last day, not when someone remembers.
Verify: net localgroup Administrators lists exactly the accounts you expect. To spot forgotten accounts, look at when each one last signed in:
Get-LocalUser | Select-Object Name, Enabled, LastLogon
Step 5: Know when something happens
Steps 1 to 4 are prevention, and prevention slips: a firewall rule gets loosened during troubleshooting, a password gets phished, a patch waits a month. You also need to hear about the events that matter while they are happening. Windows already records them in the Security log: failed sign-ins (event 4625), Remote Desktop sign-ins (4624, logon type 10), lockouts (4740), new accounts and admin-group changes (4720, 4732), new services (7045, in the System log) and a cleared log (1102).
Action: make sure those events are recorded with our copy-paste audit policy, then decide who reads them. You can script your own checks, or let a tool do it. Prixl Warden runs as a service on the server, watches those logs and emails you: immediate alerts for serious events such as password-guessing (both fast floods and slow, low-and-slow campaigns), new admin accounts or a cleared log; a daily digest at 18:00 IST; and a weekly summary every Monday. You set your office hours and it tells you about sign-ins outside them. Its security audit also checks most of Steps 1 to 4 for you.
After the five steps
Do the five steps once, verify each one, and put an hour in the calendar every quarter to check them again; access lists and firewall rules drift. Then work through the server security checklist, which adds backups with a tested restore, and read the five mistakes we see most often.
To see what your own server is facing right now, start your 30-day free trial. No card needed.

