Home › Topics › East-West Visibility & Lateral Movement › Lateral Movement Signals

Signals of Lateral Movement: Catching the Spread


Everyone worries about the break-in — the moment something gets past the firewall. But the break-in is rarely where the damage is done. One compromised laptop is an annoyance; one compromised laptop that quietly becomes ten servers over a weekend is a disaster. The distance between those two outcomes is called lateral movement: the phase where control spreads sideways from the first foothold to everything worth reaching. Catch it here, in the middle, and you turn a disaster back into an annoyance.


This guide is written entirely from the defender's chair. It does not describe how anyone moves through a network. It describes what that movement looks like from where you sit — in flow records, connection logs, and authentication events. Every signal below is something you can observe without touching the suspect host. Every one comes with the ordinary, innocent explanation as well, because most of the time the ordinary explanation is the right one. The skill is not paranoia; it is knowing which questions a signal should make you ask.


By the end you will recognize six signals. You will understand the detection logic behind each — the thresholds, the baselines, the confirmation windows. And you will have a single table for all six. For each one, it says what normally causes it, what would be worrying, and the first thing to check. Underlying all of it is one idea worth stating up front. For east-west activity, the most valuable thing you can detect is not a volume or a rate but a first time.


In this guide

  • Why "first time" is the signal that matters
  • Signal 1: new host-to-host connections between workstations
  • Signal 2: admin protocols from the wrong place
  • Signal 3: authentication fan-out
  • Signal 4: sudden internal scanning
  • Signal 5: activity at the wrong time of day
  • Signal 6: a host reaching hosts it never reached before
  • The signal table
  • Tuning without drowning
  • Where to go next

Why "first time" is the signal that matters


Normal internal traffic is boringly repetitive. The same workstations reach the same servers on the same ports every working day; the same servers depend on the same servers; the same two admin machines manage everything. A baseline — built exactly as the inter-zone monitoring guide describes — captures that repetition. Against it, spreading control stands out not because it is loud but because it is new. That could be a pair of hosts that never talked, a port never used on a pair, or a source never seen on a boundary.


This is why volume thresholds disappoint for east-west detection. Lateral movement is often quiet — a handful of small connections, nothing that would trip a bandwidth alarm. What betrays it is novelty against a well-known baseline. Tune your attention toward "have I seen this before?" and you will catch the quiet spread that a megabytes-per-second rule sails straight past. The whole rest of this guide is variations on that one theme.


Signal 1: new host-to-host connections between workstations


In almost every network, user workstations have no legitimate reason to open connections to each other. People share files through the file server, print through the print server, and collaborate through applications on servers. Two desktops talking directly — especially on a service port rather than something incidental — is unusual by design, which is what makes it such a clean signal.


The detection logic. Watch for a completed connection where both endpoints are in the user zone and the pair has not appeared in your baseline. Because same-VLAN traffic may not cross a chokepoint, this often comes from host firewall logs or access-layer flow data rather than the core. That is the reason the monitoring guide pushes you to instrument that layer. Require the connection to complete (bytes both ways), not just a lone packet, to keep the noise down.


The innocent explanation. A user running a peer-to-peer collaboration or media tool, a home-grown application that syncs directly between desktops, a game at lunchtime. These exist, which is why the signal is a prompt to check, not a verdict. The worrying explanation: one workstation reaching several others in sequence, or reaching them on administrative ports (see the next signal). That is the classic shape of a foothold looking for its next step. First check: identify both hosts in the inventory and ask whether any application legitimately connects desktops directly.


Signal 2: admin protocols from the wrong place


Some protocols exist to control a computer from a distance: remote graphical desktop, remote command shell, remote file-system and administrative shares, remote management interfaces. Named generically by function, these are remote-control protocols. Their defining property for a defender is that they should originate from a very short list of places. That means your management zone, and essentially nowhere else. An admin uses them from an admin machine or the jump host. That is the whole legitimate population.


So the signal is not the protocol itself — it is the source. A remote-desktop or remote-shell session that starts from a user workstation is worth immediate attention. So is one from a server that has no business administering other machines. So is one from any host that is not in your management zone. This is arguably the single highest-value east-west detection, because these protocols are precisely what control spreads over. Their legitimate sources are so few that the false-positive rate is naturally low.


The detection logic. Maintain the short list of hosts allowed to originate each remote-control protocol (your management zone, plus documented exceptions). Alert when one of these protocols is initiated from any source not on the list. Because the legitimate set is small, you can afford to alert on the first occurrence rather than waiting for a pattern.


The innocent explanation. It could be an admin working from their own desktop in the user VLAN instead of the jump host. (Fix the habit, and the alert stops.) It could be a help desk tool that reaches user machines from a server, or a documented management exception. The worrying explanation: any of these protocols from a source that has no administrative role at all — a reception desktop opening remote shells to servers. First check: is the source host operated by someone whose job is administration? If not, treat it as an incident until proven otherwise.


Signal 3: authentication fan-out


Spreading control usually means using credentials, and using credentials leaves a trail in your authentication logs. Those are the record your directory or servers keep of who logged in where. Two shapes in that trail are worth watching, and both are about breadth.


Fan-out is one account authenticating to many distinct hosts in a short window. A normal user logs into their own workstation and reaches a handful of servers over the day. An account that authenticates to fifteen different machines in ten minutes is behaving like something automated working through a list, not like a person. The second shape is many failures then a success: a burst of failed logins for an account or against a host, ending in a success. That is the signature of credentials being tried until one fits. The success at the end is the part that turns it from noise into an event.


The detection logic. For fan-out, count distinct destination hosts per account per short window (ten to fifteen minutes is a reasonable start). Alert when the count exceeds that account's normal breadth. For most user accounts, that breadth is very small, so the threshold can be low. Service accounts are the exception: they legitimately touch many hosts, so baseline them separately and set their thresholds higher, or you will drown. For the failure-then-success shape, watch for a cluster of authentication failures on one account or host followed by a success within the same window.


The innocent explanation. It could be a service or automation account doing its rounds. Or a user mistyped a newly changed password several times before getting it right. Another possibility is a login script touching several shares at once. The worrying explanation: a normal user account suddenly showing service-account breadth, or a failure-then-success burst on an administrative account. First check: is the account a human user or a service account, and does the set of hosts it touched match what that account normally touches?


Signal 4: sudden internal scanning


Before control can spread, whatever is spreading it usually has to find where to go — which hosts exist, which ports are open. From the defender's chair this looks like one internal host suddenly making many connection attempts across many addresses or many ports. There are lots of small, often-failing connections in a short time, a shape quite unlike normal traffic. Detecting this scanning shape is a topic in its own right. The thresholds, the horizontal-versus-vertical distinction, and how to tune it are covered in internal scanning patterns. So here it is enough to place it in the sequence. Scanning from an internal host that has never scanned before frequently comes just before the other signals in this guide. That makes it an early warning worth wiring into the same watch.


First check: confirm the source is not a sanctioned scanner. That could be a vulnerability assessment tool, a monitoring system doing discovery, or an inventory sweep. Those produce the same shape legitimately and should already be on a known list. The tell that separates the two is the source. Sanctioned scanners live in known places and run on known schedules. So a scan from a user workstation should move you from watching to acting. So should one from a server whose job has nothing to do with discovery.


Signal 5: activity at the wrong time of day


Time is one of the cheapest signals to add and one of the most revealing, because administrative activity has a rhythm. Admins patch and manage on schedules; users work business hours; backups and batch jobs run on their own clock. Against that rhythm, administrative east-west activity at an odd hour stands out. A remote-management session between servers at three in the morning, on no maintenance calendar, is worth a look. That is precisely because legitimate administration is usually a daytime, on-purpose activity.


The detection logic. Time-of-day is rarely a standalone alert — on its own it is too noisy, because plenty of legitimate automation runs at night. Its power is as a multiplier on the other signals. A new workstation-to-workstation connection during business hours earns a routine check. The same connection at 3 a.m. earns a faster one. Encode business hours and known maintenance windows, and let out-of-hours occurrences raise the priority of whatever other signal fired, rather than firing on their own.


The innocent explanation. Scheduled maintenance, overnight batch jobs, a genuinely dedicated admin, staff in another time zone. The worrying explanation: hands-on-keyboard administrative activity — interactive remote sessions, not scheduled jobs — outside any window and from an unexpected source. First check: is there a maintenance window or scheduled job that explains the timing?


Signal 6: a host reaching hosts it never reached before


This is the most general signal, and in a sense the other five are special cases of it. Every host has a habitual set of correspondents: the servers a given workstation uses, the dependencies a given server has. That set is remarkably stable. A host might suddenly begin reaching hosts well outside its usual set. A workstation that always talked to three servers might now touch a dozen across two zones. Or a server might reach peers it has no dependency on. In such cases, the shape of its conversation has changed, even if every individual destination looks ordinary on its own.


The detection logic. For each host, keep its baseline set of destinations (and the zones they fall in). Alert when a host's set of destinations expands sharply within a short window. Both the count of new destinations and the number of new zones touched matter. This is where the many-to-few shape from your baselines pays off. A client host that suddenly behaves like a management host, reaching many destinations on service ports, is the change you want to catch. Confirmation matters here. Require the expansion to persist across a window rather than firing on a single burst. Occasional legitimate spikes (a software rollout, a new application going live) happen.


The innocent explanation. It could be a genuinely new role for the host. A workstation might have just become someone's admin machine. Or a server might have gained a new dependency when an application launched. The worrying explanation: the expansion is not explained by any planned change, and it reaches into the server or management zones. First check: was there a planned change — a deployment, a new app, a re-provisioned machine — that would explain the host's new circle of contacts?


The signal table


Here is the whole set in one place. Keep it beside your dashboard; when a signal fires, the last column is where you start.


SignalNormal explanationWorrying explanationFirst check
Workstation-to-workstation connectionPeer collaboration or media tool; direct-sync appOne workstation reaching several others, or on admin portsDo the two hosts have any app that connects desktops directly?
Admin protocol from the wrong sourceAdmin working from their desktop; help-desk tool; documented exceptionRemote control from a host with no admin roleIs the source operated by someone whose job is administration?
Authentication fan-outService account doing its rounds; mistyped passwordA user account showing service-account breadth; failures then a successHuman or service account? Do the hosts match its normal set?
Sudden internal scanningSanctioned scanner, monitoring discovery, inventory sweepScan from a host that never scans, just before other signalsIs the source a known, sanctioned scanner?
Odd time of dayMaintenance window; overnight batch; off-hours adminInteractive admin session outside any window, unexpected sourceIs there a window or scheduled job that explains it?
Host reaching new hostsNew role, new dependency, planned deploymentUnexplained expansion into server or management zonesWas there a planned change that explains the new contacts?

Tip: no single signal is a conviction. Their real power is in correlation. Suppose a host scans, then opens an admin protocol to a machine it never touched. Then that machine begins scanning too. That is telling a story that no one signal tells alone. When two or three fire around the same host in the same window, stop checking and start responding.


Tuning without drowning


The failure mode of detection is not missing everything; it is alerting on so much that you stop reading the alerts. Three habits keep the signal-to-noise ratio survivable for a small team.


  • Tune for "first time," then go quiet. The first time a pair, a port, or a source appears is the alert. Once you have investigated and decided it is legitimate, add it to the baseline so it never alerts again. A detection that keeps crying about the same known-good thing trains you to ignore it — the opposite of what you want.
  • Maintain the allow-lists deliberately. These lists cover sanctioned scanners, service accounts and their normal breadth, documented management exceptions, and maintenance windows. They are what let the detections stay quiet in normal times. Treat keeping them current as part of the job, not an afterthought. An out-of-date allow-list is the usual cause of both false alarms and missed real ones.
  • Confirm before you page. Use confirmation windows (require a signal to persist or a session to complete) and correlation (two signals on one host beats one). Use them to decide what wakes someone up versus what waits in a morning review. Not every signal deserves a 3 a.m. phone call, but every signal deserves to be seen by daylight.

Above all, remember that these detections are only as good as the baseline underneath them. A signal is "different from normal," and "normal" is a thing you build and maintain, not a thing you are given. Time spent making the baseline accurate pays back every time a detection has to decide whether today is unusual.


Where to go next


You now have six signals of lateral movement described purely as things a defender observes. They include direct workstation-to-workstation connections, remote-control protocols from the wrong source, authentication fan-out, and sudden internal scanning. They also include activity at the wrong hour, and a host whose circle of contacts suddenly widens. Behind each is the same detection logic — compare against a baseline, weight by how few legitimate sources exist, confirm before acting. Behind each is also the same guiding instinct. For east-west activity, the first time something happens is the most valuable moment to notice it.


Recognizing a signal is only useful if you know what to do next. Responding to Suspicious East-West Activity covers exactly that — verify, scope, isolate, and preserve. The full topic lives on the East-West Visibility & Lateral Movement pillar page.


Put this into practice with SysPeep Monitor

Every signal here is a change against a known baseline, which means the detection is only as good as the monitoring underneath it. SysPeep Monitor keeps an always-on baseline of your devices and interfaces and discovers new hosts as they appear. It alerts on the health and reachability changes that often accompany a spread. That is the steady watch these detections sit on top of. The free trial installs on any Windows machine.

Download Free TrialLearn About SysPeep