Lateral-Movement Traffic: The Loudest Quiet Signal
If you only ever build one ransomware detection, build this one. Every stage of a ransomware intrusion leaves some footprint on the network. But most of those footprints are faint, ambiguous, or easy to confuse with ordinary life. Lateral movement is different. It is the stage where the attacker has to do something that almost nothing else on your network does. The attacker has to make a workstation behave like an administrator's console. The workstation reaches out to machine after machine over administrative protocols in a short burst of time. That is a loud, specific, structural signal. And it arrives while stopping the attack is still an ordinary bad day rather than a catastrophe.
This guide is the ransomware-specific reading of that signal. It covers the three patterns that matter most in the ransomware case. It covers the time-of-day tell that sharpens all three, and concrete thresholds set relative to your own baselines. Most importantly, it covers what to do in the sixty seconds after the alert fires. By the end you should have one alert rule you trust enough to act on immediately. You should also have a reason to act on it before you are certain.
In this guide
- Why lateral movement is the ransomware signal
- Pattern 1: admin-protocol fan-out
- Pattern 2: authentication spikes
- Pattern 3: file-share enumeration
- The time-of-day tell
- Thresholds and confirmation logic
- The minute it fires: isolate first
- The signal / threshold / first-action table
- Where to go next
Why lateral movement is the ransomware signal
Lateral movement is the general term for an intruder moving from the first machine they control to others. It happens in many kinds of intrusion. The general catalog of its signals is covered in Signals of Lateral Movement: Catching the Spread. These include first-time host-to-host connections, admin protocols from the wrong segment, and so on. This guide narrows that catalog to the ransomware case, which has a distinctive character: it is wide, fast, and administrative.
Wide, because the business model of ransomware depends on reach. Encrypting one workstation is an annoyance; encrypting every server and share at once is the event that stops a company. So the attacker wants to touch as many machines as possible, which produces fan-out — one source, many destinations. The movement is fast because once they start moving, the risk of being noticed rises with every connection. So the movement tends to be compressed into hours rather than spread over weeks. The movement is administrative because the way one machine gains control of another, at the level ransomware needs, is through the same tools you use. These are remote desktop, file sharing to administrative shares, and remote management channels. There is no mystery protocol to learn. The signal is your own admin tools, used from the wrong place, in the wrong quantity, at the wrong time.
That combination is what makes the signal both loud and quiet. Loud, because fan-out on admin ports from a workstation is structurally unlike anything users do. Quiet, because each connection looks legitimate and nobody is watching the aggregate. Your job is to watch the aggregate.
Pattern 1: admin-protocol fan-out
Start with the protocols, named by what they do rather than by number, because the number varies and the function does not. There are three families you care about:
- Remote desktop — a full interactive session on another machine's screen. Admins use it constantly. Workstations almost never originate it toward other workstations, and rarely toward more than one or two servers.
- File sharing — the protocol behind mapped drives and network shares. Every workstation talks to the file server this way. Almost none of them talk to other workstations this way, and almost none talk to a dozen servers' administrative shares in five minutes.
- Remote management — the channels that let an administrator run commands or scripts on a remote machine without a desktop session. Legitimate use comes from a small number of admin workstations, management servers, and patching systems. It comes on a schedule you can learn.
The pattern to watch for is a single source opening connections on these ports to many distinct destinations in a short window. That source is a workstation, or a single server that is not a management server. Not one connection: many, to many. The way to see it depends on what you collect. Flow records from your core switch or firewall are the ideal source. They summarize every conversation as source, destination, port, bytes, and time. A fan-out query is just "group by source and admin port, count distinct destinations per ten-minute window." If you do not have flow data, firewall logs between segments give you a partial view. In those logs, any admin-port connection crossing from a user VLAN into a server VLAN is logged. A spike of them from one address is visible. Even switch port counters help at the margins, though they cannot tell you the destination.
Here is what a fan-out looks like summarized from flow data, using a made-up workstation. Each line is one destination reached on an admin port within the same ten-minute window:
Window: 01:40 - 01:50 Source: 10.20.14.57 (WS-ACCOUNTS-12) Service Destinations Flows Bytes out remote desktop 3 3 2.1 MB file sharing 41 97 38.4 MB remote mgmt 19 19 0.6 MB Baseline for this host (30-day, same window): file sharing 1 dest, others 0
Nothing about any one line proves anything. An accounting workstation connecting to forty-one hosts over file sharing in ten minutes, when its entire history shows one, is not a thing that happens by accident. The baseline line at the bottom is what turns a number into a verdict. That is why the thresholds section below leans on it so heavily.
Pattern 2: authentication spikes
Every admin-protocol connection to a Windows machine involves a logon, and every domain logon is recorded by a domain controller. That gives you a second, independent view of the same movement. And it is often the easier one to collect, because the logs already exist on a server you already manage. There are two shapes to look for.
The first is one account, many machines. A single account logs on successfully to ten, twenty, fifty different hosts in a few minutes. Service accounts and patching systems do this on schedule; a user account, or a domain admin account outside its usual hours, does not. The count of distinct target machines per account per window is the number to watch.
The second is failures cascading, then a success. An attacker who has collected several sets of credentials but does not know which ones work will try them. The domain controller logs a burst of failed logons for one account or from one source, followed by a success. On its own, a run of failures is noise. A user who changed their password and forgot to update a saved session generates exactly this. The distinguishing feature is the success at the end. That is especially so when the successful account is not the one the source machine's user normally uses. It is especially so when the success is followed immediately by Pattern 1 from that same source. Failures then success then fan-out is the ransomware chord.
A note on where to look: authentication happens at the domain controller. But the source address in the logon event tells you which machine drove it. Correlate on that. A logon storm from a workstation using administrative accounts is Pattern 1 seen from the identity side instead of the network side. When both fire together, you are no longer looking at a maybe.
Pattern 3: file-share enumeration
Before ransomware encrypts shares it has to find them, and finding shares means connecting to them. File-server audit logs, where enabled, record every share connection and, at a finer level, every file and folder access. The pattern is one host, or one account, touching many shares. In particular, it touches shares it has never used, in alphabetical or otherwise systematic order. It reads directory listings without opening many files.
Compare that with what a person does. A user opens the two or three shares they work in, browses to a folder, opens documents, saves them. A backup job reads everything but from a known backup server on a known schedule. A search indexer reads everything but is a known service account. The enumeration signature is breadth without depth: many shares, many folders, few file reads, from a source with no history of it. On the network it shows up as many short file-sharing flows from one source to the file server in a short window. That is Pattern 1 from a different angle, with one destination but many sessions.
If you do not have file-server auditing turned on, this is one of the strongest reasons to enable it for share and directory access on your main file servers. It costs some log disk and very little performance. And it is also the primary evidence source for the encryption stage, covered in Mass Encryption Signatures.
The time-of-day tell
Every pattern above gets sharper when you add the clock. Lateral movement in ransomware incidents happens disproportionately at night, on weekends, and on holidays. That is when nobody is at the keyboard to notice a remote desktop session flicker on their screen. It is also when the IT team is least likely to be watching a dashboard. It is also when your legitimate admin activity is at its lowest. So the baseline for admin-port traffic is nearly zero and anything at all stands out.
Use this in two ways. First, run stricter thresholds outside working hours. The same fan-out count that would be a warning on a Tuesday afternoon should be a wake-somebody-up alert at 2 AM on a Sunday. Second, treat the combination of "workstation source" and "outside hours" as its own aggravating factor. A workstation that starts making admin-port connections at 1:40 AM has almost no innocent explanations left, because its user is asleep. The one legitimate exception is your own patching and management systems, and those come from known servers, which you allowlist.
Tip: the maintenance-window trap works against you here. Many teams suppress all alerts during a nightly maintenance window so that patch reboots do not page anyone. If your lateral-movement alert is inside that suppression, you have created a blind spot at exactly the hour the attacker prefers. Exempt this one rule from blanket suppression, and instead allowlist the specific management servers that generate legitimate admin traffic during the window.
Thresholds and confirmation logic
Absolute thresholds fail here. That is because "ten remote-management connections in ten minutes" is a quiet hour for your patch server and an emergency for a receptionist's laptop. The thresholds that work are relative to the source's own baseline. And they have a confirmation step so that one odd minute does not page you. Here is a starting design; tune the numbers to your network.
Metric A — distinct destinations per source per window on admin ports. For each source address, count the distinct destinations it reached over remote desktop, file sharing, and remote management in a rolling ten-minute window. Compare to that source's own typical maximum for the same time of day over the past few weeks. A useful rule of thumb: alert when the count exceeds five times the baseline and is at least ten in absolute terms. The absolute floor stops a host with a baseline of zero from alerting on its first two connections. The multiplier stops your patch server from alerting on a normal night.
Metric B — authentication events per account per window. For each account, count distinct target machines with successful network logons in a ten-minute window. Separately count failed logons from a single source. Starting rules: alert on a user account succeeding on more than five distinct machines in ten minutes. Also alert on more than ten failures from one source followed by any success on an administrative account within the same window. Service accounts get their own, learned baselines and an allowlist of expected targets.
Metric C — distinct shares per source per window. From file-server audit logs, count distinct shares connected per source in ten minutes. Alert when it exceeds the source's baseline by the same five-times-and-at-least-ten rule, excluding the backup and indexing accounts.
Confirmation logic is what keeps the rule trustworthy. Treat any single metric crossing its threshold as a candidate, not an alert. Promote it to an alert when either of these is true. The same source trips the same metric in two consecutive windows. Or two different metrics trip for the same source within thirty minutes. Then apply the time-of-day multiplier: outside working hours, a single window is enough. The sample below is a plain-language version of the rule, in the shape most alerting systems can express:
candidate = admin_dest_count(src, 10m) > max(10, 5 * baseline(src, hour_of_day))
confirmed = candidate in 2 consecutive windows
OR (candidate AND auth_spike(src OR account, 30m))
OR (candidate AND share_fanout(src, 30m))
OR (candidate AND outside_business_hours)
exclude = src in allowlist(management servers, patch server, backup server, scanner)
alert = confirmed AND NOT exclude -> severity CRITICAL, route to on-call
Two things about the allowlist. Keep it short and named — five or six specific servers, not "the server VLAN." And review it monthly, because a management server is exactly the machine an attacker would love to move from. The cleanest safeguard is a second, looser rule for allowlisted sources that fires only on a much larger multiple of their baseline.
The minute it fires: isolate first
Here is the part of this guide that matters more than any threshold. When the confirmed alert fires, isolate the source machine immediately, before you are certain. Shut its switch port, move it to a quarantine VLAN, or disable its network adapter through your management tools. Choose whichever you can do fastest and can reverse in a minute. Do not spend twenty minutes gathering evidence to be sure. Do not wait for a second opinion. Do not power it off. A powered-off machine loses the memory contents that an investigator will want. Shutting the port achieves the same containment without destroying anything.
The reason is the cost asymmetry, and it is worth internalizing because it is the opposite of your usual instinct. Your normal training says "verify before you act," because acting on a false alarm has real costs. A user loses their connection, a service goes down, and you look foolish. Here, the cost of a false positive is one user disconnected for the fifteen minutes it takes you to look at the machine and plug it back in. A false negative means waiting while a real intrusion spreads. Its cost is every machine the attacker reaches during your hesitation. At the speed of lateral movement, that can be the rest of the server room. Fifteen minutes of one person's inconvenience against days of company-wide downtime is not a close call. Isolate, then investigate.
While the port is down, the investigation is quick and mostly confirmatory. Check who is logged on to the source, what process owns the connections, and whether its destinations show the same behavior. If it was a false alarm — a technician running a script from their own workstation, say — you restore the port, note the explanation, and adjust the allowlist or threshold. If it was real, you have just moved from detection into the first hour of an incident. In that case, The First Hour: Ransomware Response for Small Teams takes over from here. The broader containment choices — proportionate, reversible steps for east-west activity that is not yet clearly ransomware — are laid out in Responding to Suspicious East-West Activity.
One more thing about the destinations. Every host the source reached on an admin port during the alert window is now a suspect. That is because lateral movement is exactly the act of turning those hosts into new sources. Your flow data or logon log gives you that list directly. Check each for the same fan-out signature, and isolate any that show it. This is the moment the "one host" cost becomes the "hosts and accounts" cost from the lifecycle diagram, and it is still cheap.
The signal / threshold / first-action table
Everything above, on one screen. The threshold column gives starting points relative to baseline; the first-action column is the same for the top rows on purpose.
| Signal | Source of data | Starting threshold | Confirmation | First action |
|---|---|---|---|---|
| Admin-port fan-out from a workstation | Flow records; inter-VLAN firewall log | Distinct destinations in 10 min > max(10, 5× baseline) | 2 consecutive windows, or any second signal, or outside hours | Isolate the source now; list its destinations |
| Admin-port fan-out from a non-management server | Flow records | Same rule, per-server baseline | Same | Isolate the server if not critical; if critical, block its admin-port egress at the firewall |
| One account, many machines | Domain controller logon events | User account > 5 distinct targets in 10 min | Pair with fan-out from the same source address | Isolate the source; disable the account |
| Failure cascade then success | Domain controller logon events | > 10 failures from one source, then a success on an admin account | Success on an account the source's user does not normally use | Isolate the source; reset the account |
| Share enumeration | File-server audit log | Distinct shares in 10 min > max(10, 5× baseline) | Breadth without file reads; not backup or indexing accounts | Isolate the source; note which shares were touched |
| Any of the above outside business hours | All of the above plus the clock | Single window suffices | Not from an allowlisted management server | Isolate and wake the on-call |
Notice that "first action" never says "investigate first." Investigation is the second action, done with the source safely off the network. If that feels uncomfortable, rehearse the isolation step on a test machine until it takes less than a minute. The rehearsal is where the comfort comes from.
Where to go next
You now have the one detection that matters most. It is admin-protocol fan-out from a source that should not produce it. It is confirmed by authentication or share-enumeration signals or simply by the hour, and answered with immediate isolation. If the alert never fires, that is a good year. If it fires and it is real, the next guide, The First Hour, is your checklist. To see what you are trying to prevent, Mass Encryption Signatures covers the stage that follows. To see where this signal sits in the whole story, start with The Ransomware Lifecycle from the Network's Point of View. The full set lives under Early Warning Signs of Ransomware on the Wire.
Put this into practice with SysPeep Monitor
The fan-out rule in this guide needs a baseline to compare against. Baselines come from history. SysPeep Monitor keeps per-device and per-interface history from SNMP and WMI collection. It has thresholds and alert rules designed to keep noise down. It routes the alerts that matter by email and other notifications. So the 2 AM workstation that suddenly starts talking to everything reaches someone who can shut the port. Free trial on any Windows machine.
Download Free TrialLearn About SysPeep