MITRE ATT&CK Enterprise · TA0008
Lateral Movement: how one compromised machine becomes the whole company
Lateral MovementTA0008
Getting in is one move. Spreading is many, and it is fast. This is the ATT&CK tactic that turns one stolen password into an entire company — and the telemetry that shows the walk.
For the owner
One stolen password should cost you one laptop. It rarely does.
What actually happens is a walk. The intruder takes over a workstation, learns the network from there, and moves to the server that holds the payroll, then to the backup machine, then to the second site. Each step is small, each step uses credentials that work, and none of them looks like an attack to a system that has been told to expect exactly this.
That is why the ransom note lists forty servers. Nobody broke into forty systems. Somebody broke into one, and then walked.
The business consequence is compounding. A single incident that touches one machine is an IT ticket. The same incident touching forty is an outage, a notification to regulators, a quarter of client trust spent, and a recovery that runs on paper checklists instead of backups. The difference between those two outcomes is not the intrusion — it is how far the intruder was allowed to travel before somebody said no.
Three questions are worth asking your team:
- If one account were misused tonight, how many machines could it reach? The answer is usually a number of shared folders, a number of remote connections, and one or two accounts that are more powerful than they look.
- Which accounts are allowed to connect from one server to another? Machines that normally reach each other are the ones a walk follows, and that map is usually more accurate than any inventory you keep.
- Who would notice a workstation logging into another workstation? Almost nothing in an office does that. It is one of the quietest signals there is, and one of the most useful.
None of this needs new software to answer. It needs the movement to be visible, and it needs it to be visible while there is still somewhere to stop.
For the analyst
The tactic
MITRE ATT&CK files Lateral Movement under TA0008. It is the tactic that turns access into reach, and the matrix files 23 techniques under it while the pinned SigmaHQ corpus carries 115 rules that name a technique of this tactic.
The tactic is narrow by design. It is not "using the network" — that is the background of every intrusion — it is the specific act of moving the foothold to another host to do the same work there. It matters more than any other mid-intrusion tactic for an arithmetic reason: it is multiplicative. One compromised account that can reach thirty machines becomes thirty compromised accounts.
| Technique | Name | Rules in the corpus |
|---|---|---|
T1021.002 | Remote Services: SMB/Windows Admin Shares | 36 |
T1021.001 | Remote Services: Remote Desktop Protocol | 15 |
T1021.003 | Remote Services: Distributed Component Object Model | 13 |
T1021 | Remote Services | 10 |
T1021.006 | Remote Services: Windows Remote Management | 10 |
T1210 | Exploitation of Remote Services | 7 |
T1021.004 | Remote Services: SSH | 5 |
T1550.002 | Use Alternate Authentication Material: Pass the Hash | 5 |
T1072 | Software Deployment Tools | 4 |
T1550 | Use Alternate Authentication Material | 4 |
T1550.001 | Use Alternate Authentication Material: Application Access Token | 4 |
T1550.003 | Use Alternate Authentication Material: Pass the Ticket | 4 |
T1570 | Lateral Tool Transfer | 4 |
T1563.002 | Remote Service Session Hijacking: RDP Hijacking | 2 |
T1021.005 | Remote Services: VNC | 1 |
T1021.007 | Remote Services: Cloud Services | 1 |
T1091 | Replication Through Removable Media | 1 |
Remote services, and how they differ
Remote Services (T1021) and its sub-techniques carry the bulk of the tactic's rules, and the sub-techniques differ in exactly the way that matters for detection: what the network looks like when the move happens.
T1021.002— SMB/Windows Admin Shares. The largest block of rules under the tactic by a wide margin, and the most important technique in it. The intruder reaches a hidden share —C$,ADMIN$,IPC$— and reads or writes files as the logged-on user, with no new credential and no new session. The network signature is a session to port 445 carrying a named pipe or a tree connect to a hidden share.T1021.001— Remote Desktop Protocol. Port 3389. A full interactive session, and therefore the most visible: the server records a type 10 logon and the client a new RDP process. The cost is that the intruder inherits the screen, the clipboard and whatever the user had open.T1021.003— Distributed Component Object Model. The quietest of the family, and the most misread. DCOM is RPC: the client talks to the endpoint mapper on port 135, which hands back a high ephemeral port, and the exchange runs over that. The named pipe is the tell —ncacn_np_— because a firewall log that records only ports shows 135 followed by a seemingly random port.T1021.006— Windows Remote Management. Ports 5985 and 5986. WinRM is PowerShell remoting: the intruder is already running commands remotely, and the command line is captured on the target host.T1021.004— SSH. Port 22. The Linux equivalent, and the one with the best telemetry:linux_auditdrecords the sshd session, the key that was used and the account, and the firewall log records the source address.T1021.005— VNC, on 5900 and above, andT1021.007— Cloud Services, where the "host" is a tenant and the move is an API call from an address that has never called before.
Exploitation of Remote Services (T1210) is the variant where the remote service itself was the weakness: an unpatched service, a published vulnerability, a management interface reachable from the network. Here the authentication is not the story — the exploit attempt is, and it shows up in process and network telemetry rather than in a logon record.
The credential problem
Use Alternate Authentication Material (T1550) covers the case where no password is used at all; the sub-techniques differ by the material:
T1550.002— Pass the Hash. The intruder extracts a hash and authenticates with it. On the wire this shows as NTLM rather than Kerberos: a type 3 logon where the service expects a type 2, from a host that does not normally produce that traffic.T1550.003— Pass the Ticket. A stolen Kerberos ticket is replayed. Nothing new happens on the wire and no new credential is used — the ticket was already valid — so this is invisible to any rule that watches authentication events.T1550.001— Application Access Token. A token minted for one application is used against another — in a cloud tenant, the difference between an API call that succeeds and one that succeeds for a token nobody expected to work there.
RDP Hijacking (T1563.002) deserves its own line: the intruder attaches to an existing remote desktop session and takes the screen, without a new session and without new credentials. No new logon event is generated at all. The only evidence is a session that was disconnected and is now connected from somewhere else.
The tools you gave them
Software Deployment Tools (T1072) is the most uncomfortable technique in the tactic, because the intruder is not using their own tooling. SCCM, Ansible, PDQ, a group policy, a script the operations team already runs — the intruder uses the mechanism that is supposed to reach every machine, and every action it takes is authorised, scheduled and logged as routine. The signal is not the tool; it is the tool under an identity that should not be running it, at an hour no deployment runs at.
Lateral Tool Transfer (T1570) is the other end of the same idea: getting the tooling onto the next host, by scp, by a copy command, or in an SMB share. Replication Through Removable Media (T1091) is the offline variant: the tooling travels on a USB stick, and there is no network log for it at all.
Where the telemetry comes from
- Endpoint logon records (
windows_sysmon). The logon type, the account inuser_name, the source address insrc_ip, the logon process inprocess_name. Type 3 is a network logon — the one that carries a remote machine — and also the most common, which is why the raw type is not the discriminator. - Linux session records (
linux_auditd). The sshd session, the account, the source address, and what ran afterwards. - Network logs (
firewall_syslog).src_iptodst_ipwithdst_port: the only source that sees the hop when the target host is not monitored. - Identity and cloud (
keycloak,m365_audit,gworkspace,cloud). The cloud analogue of a hop: a new source address for an account, an API call from an unfamiliar region.
How a rule actually reasons over this
The event that proves a hop is not rare on its own — network logons are constant. What is rare is the pair: an account, a destination, and a direction the account has never taken before.
| Signal | Columns | Why it is rare |
|---|---|---|
| Account authenticating to a host it has never reached | user_name, dst_ip | Repetition is the baseline for admin traffic; a new edge is not |
| Workstation to workstation logon | src_ip, dst_ip, user_name | Office workstations do not log into each other |
| Admin share access | file_path, user_name | A hidden share path is a deliberate choice |
| NTLM where Kerberos is expected | event_type, src_ip | Pass-the-hash leaves this signature |
| Deployment tool run by a non-deployment identity | process_name, user_name | The tool is common; the identity is not |
| Service account used interactively | user_name, event_type | Service accounts log on, they do not sit at a keyboard |
A rule that encodes the edge rather than the event looks like this:
title: Account authenticates to a host outside its established set
status: experimental
logsource:
category: process_creation
product: windows
detection:
remote_admin_share: '\\*\\C$|\\*\\ADMIN$|\\*\\IPC$'
rdp_client: 'mstsc.exe|xfreerdp.exe'
unusual_pair:
user_name|exists: true
src_ip|exists: true
dst_ip|exists: true
condition: remote_admin_share or rdp_client
fields: [user_name, src_ip, dst_ip, dst_port, process_name, host_id, event_ts]
The unusual_pair block is where the real rule lives, and it is the part usually missing. The detection above fires on the shape of the access; the rarity filter — has this account reached this host before — is what turns it from a report into an alert. Without it, every scheduled administrative task is a page.
Where analysts get this wrong
The first mistake is watching only one side of the hop. A client-side log records that the client attempted the connection; the server-side log records that it succeeded and who as. If only one of the two hosts is monitored, the hop that matters is the one you cannot see. The firewall log is the backstop for exactly this case.
The second mistake is trusting the logon type. Type 3 dominates, so a rule that fires on type 3 fires on the entire estate. The signal is the account, the destination and the direction — a workstation account reaching another workstation, an ordinary user reaching an admin share.
The third is believing that a hop always produces a new logon. T1563.002 attaches to a session that already exists. T1550.003 replays a ticket that was already valid. T1072 performs an authorised action with the deployment tool. In all three, the count of logons seen does not move, and a dashboard that only watches for new authentications shows a flat line through the whole compromise.
Detection engineering notes
- Model the graph, not the event. The natural unit here is the edge between two hosts under one account. Keep the edges, then alert on the ones that have never existed.
- Watch both directions. Logon events and network events describe the same hop from two sides, and rarely arrive together in one collection.
- Name the legitimate paths. Remote administration from a jump host, a backup server reaching the file server, the deployment tool reaching the fleet. A rule cannot tell them apart until the baseline is written down.
- Treat the deployment tool as a privileged account.
T1072is not solved by a better rule for the tool; it is a question of which identities are allowed to run it.
Where DetectMonkey fits
Lateral movement is a race, and the race is measured in minutes: the gap between the first hop and the second is often shorter than the time it takes to walk to the next desk.
DetectMonkey is built for that race. The collectors read the authentication and connection telemetry a walk leaves behind — the source address, the destination port, the account, the logon record — and the pinned corpus of 3164 Sigma rules evaluates them as the events arrive rather than after they are indexed, so the second hop is alerted on while the third is still being planned. Because the last 72 hours of telemetry stay queryable, an investigation can start from the machine that was left and ask, in one query, which other machines that account reached inside the same window.
The platform detects and alerts. It does not isolate, quarantine or block anything on your systems, and it does not change your configuration. The telemetry is processed and stored on private servers operated by Codein in New Brunswick, Canada, not in a public cloud, and analysts work in English and French. Early access is 90 days at no charge, for up to 100 seats, founder-led onboarding and no card required.
See it on your own telemetry
Early access is 90 days at no charge, for up to 100 seats, founder-led onboarding, no card required.