DFIR Group Project · Case File
Catching an Intruder Three Ways
Multi-Source Detection & Forensic Reconstruction of a Simulated Windows 11 Intrusion Using Open-Source Tooling
- Endpoint Logs
- Disk Artifacts
- Physical Memory
One attack, three independent lenses — and no single source saw all of it.
01 Overview
The research question
Incident responders rarely get the whole story. They reconstruct an attack from fragments: an event log here, a prefetch file there, maybe a memory image if they were quick. Every source is incomplete in its own way. We wanted to measure how incomplete, by running a known attack and checking what each lens could recover.
When a known set of attacker techniques is executed, how much can be detected and forensically reconstructed from each evidence source — and where does each one succeed or fail?
- 11ATT&CK techniques attempted
- 7ran with full success
- 8/8executed techniques detected
- 297/297evidence files hash-verified
02 Methodology & Safety
A five-step pipeline
-
Isolate
Windows 11 25H2 VM on VMware, host-only networking.
-
Instrument
Sysmon + PowerShell script-block logging.
-
Attack
11 ATT&CK techniques via Atomic Red Team.
-
Acquire
Memory, logs, prefetch, registry, all SHA-256 hashed.
-
Analyse ×3
Detection, Disk and Memory tracks, run independently.
03 Tools
Open-source, and chosen for a reason
| Tool | Role | Why this one | Alternative considered |
|---|---|---|---|
| VMware Workstation | Runs the isolated victim VM | Host had Hyper-V active; VMware cooperates where VirtualBox crawls | VirtualBox (slow under Hyper-V) |
| Sysmon | Rich endpoint telemetry | Full command lines, parent, hashes — beyond default logs | Default Windows auditing |
| Atomic Red Team | Safe adversary emulation | One ready test per ATT&CK technique; no real malware | Caldera / manual exploitation |
| Hayabusa + Chainsaw | Detection over logs | ~4,500 Sigma rules applied offline in seconds; no SIEM to deploy | Full Wazuh/Splunk SIEM |
| Prefetch + Registry (Eric Zimmerman tools) | Disk forensics | Durable execution & persistence proof that survives reboots | — |
| Volatility 3 | Memory forensics | Standard open-source memory framework | WinPMEM / vmss2core (acquisition) |
04 The Attack
Eleven techniques, three outcomes
| ID | Technique | Tactic | Result |
|---|---|---|---|
| T1059.001 | PowerShell Command Execution | Execution | OK |
| T1053.005 | Scheduled Task (persistence) | Persistence | OK |
| T1547.001 | Registry Run Key (persistence) | Persistence | OK |
| T1112 | Modify Registry | Defense Evasion | OK |
| T1016 | Network Config Discovery | Discovery | OK |
| T1082 | System Information Discovery | Discovery | OK |
| T1057 | Process Discovery (tasklist) | Discovery | OK |
| T1033 | System Owner/User Discovery | Discovery | Partial |
| T1047 | WMI Reconnaissance | Discovery | Failed wmic removed in Win11 25H2 |
| T1070.004 | File Deletion | — | Skipped no Windows atomic |
| T1057 (test 1) | Process Discovery | — | Skipped no Windows atomic |
Note: T1047 and the WMIC part of T1033 both failed because Microsoft removed wmic in Windows 11 25H2. Attacker tooling has to keep up with OS changes too.
05 The Coverage Matrix
Who saw what
Each of the 8 executed techniques, checked against each evidence source. Hover, focus or tap a cell for the detail.
| Technique | Detection logs | Prefetch | Registry | Memory |
|---|---|---|---|---|
| T1059.001 PowerShell | ||||
| T1053.005 Scheduled task | ||||
| T1547.001 Run key | ||||
| T1112 Modify registry | ||||
| T1016 Net discovery | ||||
| T1082 System info | ||||
| T1057 tasklist | ||||
| T1033 Owner/user (partial) |
- Detected / recovered
- Detected with a caveat
- Not present
- Not applicable (benign default)
Every executed technique was recovered by at least two independent sources, and 7 of 8 by both logs and prefetch. Memory alone caught nothing.
06 Results
Three tracks
Track 1 · Detection over logs
- 10,277events processed by Hayabusa
- 5,608rule hits (~50% data reduction)
- 8/8executed techniques detected
- 2custom Sigma rules written
Hayabusa ran the Sigma corpus offline over the Sysmon and PowerShell logs. Chainsaw independently confirmed the results. All 8 executed techniques fired at least one rule. Two of them came with caveats:
- T1057 missed by default rules. The
tasklistdiscovery wasn't flagged by the stock corpus. We only caught it with threat-hunting and a custom Sigma rule. - T1112 mis-tagged as T1137. The detection fired, but under the wrong ATT&CK ID. A defender triaging by ID would attribute it to the wrong technique.
Track 2 · Disk artifacts
- 0–4 sprefetch offset from ground truth
- 7 dayspersistence still live afterwards
- 0WMIC.EXE prefetch entries
- Prefetch timed every technique. It placed each technique's binaries within 0–4 seconds of ground truth.
arp,netsh,nbtstat,systeminfoandtasklisteach ran exactly once, so those runs can only be the attack. - Negative evidence. There is no
WMIC.EXEprefetch at all. The absence independently confirms that the WMI techniques failed. - Persistence outlived the attack. Seven days later, an HKCU Run value named
Atomic Red Teamand two scheduled tasks (OnLogon / OnStartup, actioncmd.exe /c calc.exe) were all still live. - T1112 left no registry trace. It rewrote a value to its existing default, so only Prefetch can prove it ran.
Track 3 · Physical memory
07 Errors We Solved
What broke, and what we did
-
ProblemHyper-V blocked VirtualBox
FixSwitched to VMware Workstation.
-
ProblemWindows 11 refused to install (4 GB RAM / 52 GB disk minimums)
FixResized the VM to 4 GB / 60 GB.
-
ProblemDefender blocked our PowerShell (Tamper Protection,
0xc0000142)FixDisabled real-time protection and added an exclusion via the Security GUI.
-
ProblemWindows updated mid-setup
FixRe-verified timezone, Sysmon and Defender state before attacking.
-
ProblemWinPMEM driver blocked (
0x241)FixFell back to a VMware suspend image and documented it.
-
ProblemVolatility couldn't parse the suspend image
FixReported it as a modern-OS acquisition finding.
08 Team & Learnings
Who did what, and what we learned
-
Lavanya Sharma
Lab build, instrumentation, attack execution, evidence acquisition, memory track & integration
-
Ayushi
Detection: Hayabusa, Chainsaw, custom Sigma
-
Mansi Sharma
Disk forensics: prefetch & registry, Autopsy validation
-
Alina Noorain
Memory track (initial)
- No single source is complete.
- A detection that fires is not necessarily a correct detection.
- Persistence lives on disk long after an attack; memory is fleeting.
- On a hardened OS, acquisition is the hard part.
- Upfront instrumentation decides what evidence exists at all.
- Documenting failures honestly makes the work stronger.