In this section

Windows Initial Access and Lateral Movement Forensics: Email, Browser, 4648 and PsExec

Module 0

Introduction

Every intrusion has a first host and, usually, a second.

The first question an investigator is asked is how the attacker got in, and the second is where they went next, because the answer to the second decides how many machines need collecting and how far the incident reaches. Both are answered from one host's evidence more completely than most people expect, and both have a limit that a single host cannot get past.

Email01:47:11SPF and DMARC failLink01:49:22clicked, not typedSign-in page01:49:44lookalike domainDownload01:50:334.3 MB fileRuns01:51:22Edge its parent

The way in on 18 March, from three different sources: the mailbox, the browser profile and Sysmon. Four minutes separate the email from the program running.

This lesson reads the way in and the way across on this course's ransomware case: a finance workstation, a phishing email, and an attacker with an administrator's password reaching for the file server and the domain controller.

Module 4 teaches each part in depth: finding patient zero, how a file's arrival is recorded, authentication as movement, Remote Desktop, what each remote execution technique leaves, pass-the-hash, and choosing which host to collect next.

01

Where the Way In Is Written

Three sources, one timeline

Initial access is rarely recorded in one place. The ransomware case's way in is spread across the user's mailbox, the browser profile and the Sysmon log, each kept by a different program with its own format, and none of them knows about the others. The chain above came from putting the three on one clock.

The mailbox says what arrived and when. The browser profile says what the user did with it: which links were followed, which pages loaded, what was downloaded and from where. The process log says what ran, from what parent, and what it connected to.

Read alone, each is ambiguous. An email with a suspicious link might never have been opened, a download might never have run, and a process might have come from anywhere. Read together, they make a sequence with no gaps.

The attacker's whole way in took four minutes from the email's arrival to the program running, and the order is the evidence. A program that runs four minutes after a lure arrives, from the browser that visited the lure's domain, with a connection back to the address that sent it, is not three coincidences.

Before reading any of them, it helps to know what each source is and what keeps it:

The way in: three sources and what each holds

Join on the clock

Mailbox

what arrived, from where, and its authentication results

Kept by

the mail client's profile, or the server's audit log

Browser

pages visited, how each visit began, files downloaded

Kept by

History database in the user's browser profile

Process log

what ran, its parent, its account, its connections

Kept by

Sysmon, or Security event 4688 where auditing is on

The last row has an alternative because not every host runs Sysmon. Where it is absent, the Security log's process creation event, 4688, records the same launch if process auditing was enabled, and Module 4 reads both. Where neither exists, the execution artifacts from lesson 0.2 still prove the program ran, without its parent or its connections.

The records below are the parsed exports the course uses: the mailbox export from the user's mail profile, Edge's History database read with a browser parser, and Sysmon read with EvtxECmd. Module 4 reads each from its raw source.

02

The Email

What the header says before anyone opens it

The mailbox export holds thousands of messages. Most are ordinary, and finding the one that matters starts with the authentication results the receiving server wrote into the header:

Subject       DocuSign: NE Contract Amendment - Review Required
From          docusign@d0cusign-secure.com
Return-Path   bounce@d0cusign-secure.com
Received      from mail-out.d0cusign-secure.com (203.0.113.44) by mx.northgateeng.com 
SPF           fail (domain d0cusign-secure.com does not designate 203.0.113.44 as perm
DKIM          none
DMARC         fail (p=quarantine; dis=quarantine)
Date          2026-03-18T01:47:11Z
To            t.bennett@northgateeng.com

Three checks failed or were absent. SPF failed: the domain's published record does not authorize 203.0.113.44 to send its mail. DKIM is none: the message carries no signature at all. DMARC failed, and the domain's own policy asks receivers to quarantine it. The sender's domain, d0cusign-secure.com, uses a zero where the real brand has a letter o, and the Return-Path is on the same lookalike domain.

None of that proves the user opened it. A header is written when a message is received, not when it is read, and the export shows it arrived at 01:47:11 UTC, in the middle of the night. What makes it the way in is what happened on the host in the minutes after, which the next two sources record.

Mail that fails DMARC with a quarantine policy should normally land in a junk or quarantine folder rather than the inbox, and the export does not record which folder this message was delivered to. Whether the receiving system honored the policy is a question for the mail platform's own logs, and a gap worth raising with whoever runs it, because it decides whether other staff received the same message.

The address in the first Received line is the one to keep. Header fields such as From are written by the sender and can say anything; the Received line nearest the recipient's own server is written by that server, recording the address that actually connected to it. It is the most trustworthy field in the header, and here it is 203.0.113.44.

Lookalike domains are worth checking character by character rather than by eye. A zero for an o, an rn for an m, or an extra word such as secure added to a brand are the common patterns, and they are designed to pass a glance.

The Return-Path, where bounces go, is set by the sending system and is usually on the attacker's own domain even when the display name is perfect. Here every address in the header is on the same lookalike domain, which is the sign of infrastructure built for the purpose.

03

The Browser

A link followed, a page loaded, a file saved

Edge keeps a History database in the user's profile: every page visited with its URL, title, visit count and the way the visit started, and every download with its source and destination. Filtered to the lookalike domain:

2026-03-18T01:49:22  history   https://d0cusign-secure.com/review?token=a8f7e6d5c4b3&d
2026-03-18T01:49:44  history   https://d0cusign-secure.com/auth?redirect=review
2026-03-18T01:50:33  download  https://d0cusign-secure.com/download/DocuSign_Review.ex
                     saved to  C:\Users\t.bennett\Downloads\DocuSign_Review.exe
                     bytes     4,523,891
                     referrer  https://d0cusign-secure.com/review

The two history rows are a link from the email and a sign-in page on the same domain. Both have a transition type of link and a typed count of zero, which means the user arrived by clicking rather than by typing an address. The sign-in page is a credential harvesting page dressed as the real brand's, and anything the user typed into it went to the attacker.

The download row is the payload: a 4,523,891-byte executable saved to the Downloads folder at 01:50:33, with its source URL and the page that referred to it. The referrer ties the download to the review page the email linked to. A file in Downloads can arrive many ways; this one has a documented path from a link to a page to a saved file.

The transition type is worth knowing as a field. Browsers record how each visit began: a typed address, a bookmark, a link, a redirect, a form submission.

A typed visit to a lookalike domain would mean the user went there deliberately; a link visit means they followed something, and with the email four minutes earlier, the email is the obvious candidate. The browser does not record which application the link was clicked in, so that last step is inference from timing and content, and is written as such.

Browser history has well-known limits. Users and attackers can clear it, private browsing does not write it, and only the most recent visit time is kept per URL in this table. Module 4 and lesson 0.10 read the other places a browser records the same visits, which is how the gaps are filled when one is missing.

04

The First Program

It ran, and it called home

The download is only a file until something runs it. Sysmon records the moment it became a process, and the first thing it did:

01:51:22  Event 1 C:\Users\t.bennett\Downloads\DocuSign_Review.exe
          parent  C:\Program Files (x86)\Microsoft\Edge\Application\msedge.exe
          user    NORTHGATEENG\t.bennett, integrity Medium
01:51:33  Event 3  DocuSign_Review.exe
          to      203.0.113.44:443 tcp

The parent is msedge.exe: the user ran the file from the browser's download bar, under their own account, at medium integrity, the level an ordinary user's programs run at. Eleven seconds later the new process opened a connection to 203.0.113.44 on port 443.

That address is the one the email header recorded as the sending server. The same infrastructure that delivered the lure received the first connection from the payload, which closes the chain from the outside: the email is no longer merely suspicious and close in time, it is linked to the program by an address that both records hold independently.

The integrity level matters for what comes next. A medium-integrity process can read and write the user's own files and registry hive but cannot change system settings, install services or read other users' data.

Everything the attacker did afterward needed more than that, which is why the next record shows the process tree jumping to SYSTEM: the payload either carried an exploit or abused a misconfiguration to raise its own privilege. Module 4 reads the common escalation routes and what each leaves behind.

This is the point at which the investigation changes question. Up to here it was how the attacker got in; from here it is what they did with the foothold, and lesson 0.2 already showed the four programs that ran next. The one that matters for moving on is the second, because of what it was for.

05

From a User to an Administrator

The account changes

An attacker who lands as an ordinary user can read that user's files and little else. To reach other machines they need an account those machines trust, and the next records show them getting one:

02:01:15  Sysmon 1  mshelper64.exe --dump  as NT AUTHORITY\SYSTEM
02:08:22  4624      type 10  NORTHGATEENG\nge-admin
                    source 127.0.0.1, elevated token Yes

At 02:01:15 a program named mshelper64 ran as SYSTEM with the argument --dump, the shape of a credential-dumping tool reading passwords and hashes out of memory. Seven minutes later the Security log records a type 10 logon, a Remote Desktop session, for nge-admin, a domain administrator, with an elevated token.

The source address of that logon is 127.0.0.1. Microsoft's documentation for the logon events notes that 127.0.0.1 means localhost: the Remote Desktop connection came from this machine itself, not from a client somewhere else on the network. An administrator sitting at their own desk and connecting to this workstation would have produced their own address.

A loopback source is what a session looks like when it arrives through something running on the host, such as a tunnel an attacker set up, and it is one of the indicators Module 4 reads in detail.

The elevated token field adds a second detail. Microsoft records whether the session received a token with administrative rights in effect; for an ordinary user it is No, and for nge-admin it is Yes. The session that opened at 02:08 could do anything an administrator can do on this host, without a further prompt.

The order is what makes the finding. Credentials were dumped at 02:01; an administrator's account was in use at 02:08; the user whose machine it is had no administrator rights. The account change is the attacker's, and every action from here on carries an administrator's name.

06

The Way Across

Three witnesses to two moves

With an administrator's password, the attacker reached for two more machines. The source host recorded those attempts in three separate places:

Security log4648, 02:12explicit credentials for NGE-FILESERVER4648, 02:14explicit credentials for NGE-DC01SysmonEvent 1, 02:12PsExec \\NGE-FILESERVER -u nge-adminEvent 1, 02:14PsExec \\NGE-DC01 -u nge-adminNTUSER.DATServers keyNGE-FILESERVER, hint nge-adminServers keyNGE-DC01, hint nge-admin

Two moves, three witnesses on the source host. Each lane records the same two attempts in a different way, and none of them can say whether either attempt worked.

The Security log's event 4648 is the most specific. Microsoft describes it as generated when a process attempts a logon by explicitly specifying an account's credentials, and it records the target server's name and the full path of the process that did it:

02:12:11  4648  nge-admin -> nge-admin on NGE-FILESERVER
          process C:\Users\t.bennett\AppData\Local\Temp\PsExec.exe
02:14:11  4648  nge-admin -> nge-admin on NGE-DC01
          process C:\Users\t.bennett\AppData\Local\Temp\PsExec.exe

Both attempts came from PsExec in the user's Temp folder, using nge-admin's credentials, toward the file server and then the domain controller. Sysmon's process creation events record the same two launches with their full command lines, including the -u switch that PsExec's documentation describes as the user name for login to the remote computer. The password is masked in the log.

The third witness is the user's own registry hive. The Remote Desktop client remembers the servers it connected to and the user name it offered each one:

2026-03-18T02:14:33  Terminal Server Client\Default
                     MRU0=NGE-FILESERVER  MRU1=NGE-DC01
2026-03-18T02:12:33  Terminal Server Client\Servers\NGE-FILESERVER
                     UsernameHint=NORTHGATEENG\nge-admin
2026-03-18T02:14:33  Terminal Server Client\Servers\NGE-DC01
                     UsernameHint=NORTHGATEENG\nge-admin

The client's MRU list names the same two servers, and each server's key holds nge-admin as the user name hint. Microsoft also calls 4648 a routine event that occurs periodically during normal operation: scheduled tasks and the RUNAS command produce it, and on a busy server there are many.

What makes these two stand out is the combination of fields rather than the event itself: a domain administrator's credentials, supplied by a program in a user's Temp folder, toward servers that a finance workstation has no business administering.

Three independent records, written by the operating system, Sysmon and the Remote Desktop client, agree on which machines the attacker reached for and with which account.

07

What Only the Targets Can Say

Intent on one host, outcome on another

Every record in the last section is written by the source host, and the source host knows what it tried. Whether the file server accepted the connection, and what ran there afterward, is written on the file server. Whether the domain controller authenticated the account is written on the domain controller:

1. Source hostwhat was launched, with which account, toward which server2. Networkthat a connection was made, how long, how much data3. Target hostwhether the logon succeeded, and what ran there afterward4. Domain controllerthe authentication, its result, and any lockout

Where each half of a lateral movement is written. A source host shows intent and method; only the destination and the domain controller show the outcome.

The source host does hold one record about the account's fate:

Time            2026-03-18T02:14:33Z
Event           4625, logon type 3
Account         NORTHGATEENG\nge-admin
Source address  10.1.10.10
Failure reason  Account locked out
Status          0xC0000234

At 02:14:33, a network logon to this workstation as nge-admin, from 10.1.10.10, failed because the account was locked out. Microsoft's documentation gives 0xC0000234 as the typical status for exactly that failure. By that moment, the administrator account the attacker had been using was no longer usable anywhere in the domain, whatever had happened at the two targets before it.

The failed logon is itself a lead. Something at 10.1.10.10 tried to use the administrator account against this workstation at the moment it was locked out, and nothing on this host says what that machine is or why it tried. It might be the attacker on a second foothold, an automated tool reusing the stolen credentials, or a legitimate system that happened to authenticate at that minute.

The address goes on the list of hosts to identify, alongside the two named targets, and the lockout itself goes to whoever runs the domain controller, whose own log records which attempts tripped it.

That still does not say whether the move to the file server succeeded in the two minutes before the lockout. The case brief records that network segmentation stopped the ransomware spreading, but this host's evidence cannot confirm or deny a session on the file server. The finding is written to that limit:

Lateral movement: what this host can answer

Scope the next collection

Launched?

yes: Sysmon command lines and 4648, 02:12 and 02:14

With what?

PsExec from the user's Temp folder

As whom?

nge-admin, credentials given on the command line

Toward?

NGE-FILESERVER, then NGE-DC01

Succeeded?

not answerable here: needs the targets' logs

Next step

collect from both targets and the domain controller

The case's own evidence set is a single host by design, which is the common situation at the start of a real investigation: one machine alerted, it was imaged, and everything else is a question. Writing the finding with its limit is what tells the incident lead which machines to image next and why.

The last row is the practical consequence. A lateral movement finding from the source host is a list of the next machines to collect from. An investigator who stops at the first host has found where the attacker went, but not what they did when they got there.

08

What Module 4 Builds On It

Routes in, routes across, and the map

Module 4 reads each part of this lesson in depth, and adds what a single host cannot teach: how to tell whether the host that alerted was really the first, how to order an unbounded queue of hosts to collect, and how to write a chain that has a gap in it.

The order Module 4 takes:

Module 4, in order

Module 4

Patient zero

the first host, and why the alerting one rarely is

Arrival

Mark of the Web and the delivery chain to execution

Movement

authentication as movement; RDP from both ends

Techniques

which host holds what for each one; pass-the-hash

The queue

choosing the next host, collecting from a live estate

Gaps

a chain with a gap, and when to stop

The first two rows extend the way in. The alternate data stream Windows attaches to downloaded files records where a file came from, which would have tied DocuSign_Review.exe to its URL even with the browser history cleared. The middle rows extend the way across, and its lesson on remote execution techniques is organized by exactly the question section 7 raised: which machine holds the evidence for each technique.

The last two rows are about the investigation itself: every host examined names more hosts, the queue grows faster than it can be collected, and the order decides how much evidence survives. A chain that ends where evidence ran out is still a finding, provided it says which kind of absence it hit.

The habit to carry into it is to read the way in as a chain with no gaps, each link held by an independent record, and to read the way across as two halves. Intent and method are on the source host and can be established from it. Outcome is on the destination and the domain controller, and a finding about it waits until their evidence is collected.

The investigation module later in the course reads this case end to end; lesson 0.5 reads what the attacker left behind on YORK-WS-033 to run again, which is where a service named after PsExec appears.

Do

Practice

How did they get in, and where did they go?
prepared evidence

The prepared evidence below is the hour before encryption on YORK-WS-033: recent messages in the user's mailbox, Sysmon's process and network events, and the Security log's events for the administrator account. Find the message, tie the program to its sender, and pick out the attempts on other machines.

If you have a Windows machine of your own, filter your Security log for event 4648 and read the process names and target servers it records. Most will be ordinary: scheduled tasks and saved credentials. Knowing what ordinary looks like on your own machine is what makes PsExec stand out on someone else's.

The way in is read across the mailbox, the browser and the process log, put on one clock and joined by an address that two of them hold independently. The way across is read on the source host as intent and method: which tool, which account, which server. Whether it worked is written on the targets, and the finding says so.

The next lesson reads what an attacker leaves behind so that their access survives a reboot or a password change.

Resources

Microsoft revises these pages; where a page and this lesson disagree, the page is newer.