Why is a Windows server answering Remote Desktop on TCP 3389 from the internet treated as an incident route, and what do you fix first?
answer
- Anyone can start a connection
- Two routes: flaw and guessing
- Take it off the internet first
- Authenticate before a session exists
basics
~20 sAn internet-reachable RDP listener lets strangers try pre-authentication flaws such as BlueKeep and guess passwords around the clock. Remove direct exposure with a gateway or VPN, require NLA and multifactor authentication, and keep the host patched.
solid answer
~40 sRDP on TCP 3389 is a full logon service. Open to the internet, it gives strangers two routes in: **pre-authentication flaws** in Remote Desktop Services (BlueKeep, CVE-2019-0708, is the model, and the records for CVE-2019-1181, CVE-2019-1182, CVE-2019-1222 and CVE-2019-1226 repeat the pattern) and **password guessing**, which shows up as bursts of event 4625. Fix in this order: take the listener off the internet by removing the router forward or any-source firewall rule, and put a gateway or VPN in front of it; require Network Level Authentication so a session request is authenticated first; add multifactor authentication at the gateway; apply updates; and set an account lockout threshold. Moving the service to another port is not on the list, because the same code still answers.
go deeper
Recall that 3389 is a logon service reachable by strangers when exposed, and that the first fix is to remove the direct exposure, not to rename the port.
Explain the two routes, a pre-authentication flaw and password guessing, and name the layered fixes: gateway or VPN, NLA, multifactor authentication, patches, lockout.
Show the order and the costs: perimeter blocks break legitimate access, lockout can be turned against staff, and a gateway becomes the next patching target.
Frame it as a standing rule: no host answers RDP to the internet, with a short exceptions register and evidence that the rule holds.
## What an exposed 3389 listener offers Remote Desktop Protocol (RDP) is the protocol behind Windows Remote Desktop. A server running Remote Desktop Services listens on **TCP 3389** by default, and anything that can reach that port can start a connection and meet the service's code before anyone has shown who they are. When the port is reachable from the whole internet (a router port forward, a cloud rule that allows any source, a firewall exception nobody removed), every host on the internet can try, all day, at no cost to them, and nobody on your side notices unless you log it. ## The two routes in | Route | Why it works | How it shows on your estate | |---|---|---| | Pre-authentication flaw in Remote Desktop Services | The vulnerable code is reached by a connection request, before any logon. BlueKeep (CVE-2019-0708) is the model; the records for CVE-2019-1181, CVE-2019-1182, CVE-2019-1222 and CVE-2019-1226 describe the same shape: "pre-authentication and requires no user interaction" | An unpatched release on the affected list; possibly no failed logon at all | | Password guessing at the logon | The service accepts logon attempts from any source, so weak or reused passwords can be tried again and again | Runs of event 4625 failures, and lockouts (status 0xC0000234) of real staff | Two consequences follow. Patching closes the first route only for flaws already known, and strong passwords address only the second; neither turns a world-reachable listener into a sound design. And a quiet log is not proof of safety, because the first route does not need a failed logon. ## Why reviews call it an incident route CISA's alert on BlueKeep (AA19-168A) says the flaw lets an attacker run code on an unprotected system, add accounts with full user rights, view, change or delete data and install programs, with no user interaction and before authentication. It compares the potential spread with the WannaCry attacks of 2017. Incident reviews commonly name internet-reachable remote access as an entry point, ransomware included, but none of the sources behind this answer measure how often, so treat that as common experience and not as a figure. ## The first fixes, in order 1. **Take the listener off the internet.** Remove the router forward or the any-source firewall rule. Microsoft's page on reaching a PC from outside its network says that opening the port to the internet "isn't recommended", prefers a VPN, and notes that routers can limit which source addresses a forward accepts. 2. **Give legitimate users another path.** A gateway that carries RDP inside HTTPS, or a VPN, so the session host has no inbound exposure of its own. The gateway is its own subject, with its own patching and policies. 3. **Require Network Level Authentication (NLA).** A session request must then be authenticated before a session exists. CISA names this as a mitigation for BlueKeep on Windows 7 and Windows Server 2008 and 2008 R2. 4. **Add multifactor authentication at the gateway.** Microsoft's planning page says that even if passwords are compromised, the extra verification prevents unauthorized access. 5. **Patch**, and retire releases that no longer receive updates. 6. **Set an account lockout threshold** and watch event 4625, accepting the trade it brings (covered with lockout). 7. **Open the port only when needed**, for the rare path that cannot use a gateway, and close it again. ## What is not on the list Changing the listening port. Microsoft documents how: the `PortNumber` value under the `RDP-Tcp` key, plus a new firewall rule for the new number. The page offers it "for security or configuration purposes" and does not say it removes exposure. The same protocol, with the same code and the same flaws, answers on the new port; only clients must now name it. It is a configuration option, not a control, and an external check of your own ranges that tests every port still finds it. ## Costs and residual risk - Blocking 3389 at the perimeter stops legitimate remote administration too, and, as CISA notes, may not prevent unauthenticated sessions from being started inside the network. - A gateway or VPN becomes the new internet-facing service and needs its own patching. - NLA reduces who can reach the flaw; it does not replace the update. - A lockout threshold turns guessing into a possible denial of service against staff.
- Why is changing the listening port from 3389 to another number not a control?Microsoft's page offers the change for security or configuration purposes and documents only the PortNumber value and the matching firewall rule. The same protocol and the same Remote Desktop Services code answer on the new port, so a pre-authentication flaw stays reachable and guessing still works. It only changes what a client must type, and an external check of your own ranges that tests all ports still finds it.
- Does blocking TCP 3389 at the perimeter finish the job?CISA lists it as a BlueKeep mitigation but states two costs: it blocks legitimate RDP sessions, and it may not prevent unauthenticated sessions from being initiated inside a network. So it removes the internet route, not a host's exposure to a compromised neighbour. Patching, NLA and restricted logon rights still apply.
- If a user must reach a PC from outside the network, what does Microsoft's guidance prefer?A VPN, so the PC is not opened to the public internet. If a router forward is used anyway, the page points out that most routers can limit the source address or network that may use the mapping, which avoids exposing the port to everyone. It still leaves the listener reachable from the allowed range.
Moving the listening port is like repainting the back door of a shop that opens onto the street: anyone who looks still finds it, and the real fix is to stop the door facing the street.
- A strong password makes an internet-facing 3389 safe
- Changing the port number removes the exposure
- Failed logons from the internet are harmless background noise
- NLA means the host no longer needs patching
- Blocking 3389 at the perimeter also stops anyone already inside
- A quiet event log proves nobody reached the service