
Overview
A phishing email with a zipped "invoice" gave an attacker a foothold on michael.ascot's workstation (win-3450). In under four minutes of hands-on activity, the attacker stole client and investor documents from the finance file share over DNS. The real intrusion was buried among 28 alerts, most of them noise from two untuned rules.
The attacker's objective was data theft: copying sensitive financial documents (a client portfolio summary and an investor presentation) out of the organisation. The likely uses are extortion, fraud or sale.
| Count | |
|---|---|
| Alerts closed | 28 |
| True positives (one incident on win-3450) | 15 |
| False positives (spam and benign Windows processes) | 13 |
| Classified incorrectly | 2 (alerts 1000 and 1010, both marked true positive but actually benign) |
The network picture is simple. Everything happened on one internal host (win-3450), which reached one internal file server (FILESRV-01) and one external attacker domain (haz4rdw4re.io).
Attack chain timeline
From phishing email to finished exfiltration took 25 minutes. The hands-on stage, from tool drop to the last DNS query, took under 4 minutes. Every step after delivery ran from one PowerShell process (PID 3728, with PID 9060 dropping the tool) in michael.ascot's Downloads folder.
| Time (16 Nov 2026) | Stage | What happened | Where | Alert | MITRE ATT&CK |
|---|---|---|---|---|---|
| 08:17:11 | Initial access | Fake overdue-invoice email with ImportantInvoice-Febrary.zip | External sender → michael.ascot's mailbox | 1005 | T1566.001 |
| 08:38:41 | Tooling | PowerView.ps1 written to Downloads | win-3450 | 1020 | T1105 |
| 08:40:36 | Discovery and collection | Finance share mapped as Z: | win-3450 → FILESRV-01 | 1022 | T1135, T1039 |
| 08:41:23 | Staging | Robocopy copies the share into the exfiltration folder |
FILESRV-01 → win-3450 | 1023 | T1074.001 |
| 08:41:34 | Defence evasion | Z: mapping deleted | win-3450 | 1024 | T1070.005 |
| 08:42:21 | Exfiltration | ZIP of two files sent in 8 DNS queries | win-3450 → haz4rdw4re.io | 1025–1032 | T1560, T1048 |
| 08:42:37 | Exfiltration | THM{} string sent in 2 DNS queries | win-3450 → haz4rdw4re.io | 1033–1034 | T1048 |
Initial access and tooling (1005, 1020)
The intrusion started with a fake overdue-invoice email carrying a zip, and the first sign of compromise was an offensive recon script appearing 21 minutes later.
1005: Suspicious attachment (true positive)
An external sender, john@hatmakereurope.xyz, emailed michael.ascot@tryhatme.com at 08:17:11 with the subject "FINAL NOTICE: Overdue Payment - Account Suspension Imminent" and the attachment ImportantInvoice-Febrary.zip.
The lure had every classic sign: an unfamiliar .xyz sender posing as a supplier, a threat of suspension "today" and legal action within 24 hours, an instruction to open the attachment "immediately", an "invoice" delivered as a zip instead of a PDF, and a typo in the filename. Motive: get the user to run the payload hidden in the zip.
When I first triaged this alert, there was no sign of execution. The later alerts proved the attachment was opened, so it belongs in the escalated incident.
1020: PowerView.ps1 in Downloads (true positive)
At 08:38:41, PowerShell (PID 9060) wrote PowerView.ps1 to C:\Users\michael.ascot\Downloads\. PowerView is a PowerSploit reconnaissance script for enumerating Active Directory and finding network shares.
The alert was rated Low, and the rule name sounds routine. The context made it a clear true positive: an attacker tool, on the phishing victim's host, two minutes before the finance share was mapped. Motive: discover where valuable data lives on the network, which led straight to \\FILESRV-01\SSF-FinancialRecords.
Escalation: required for both, as part of the win-3450 incident.
Collection (1022, 1023, 1024)
In 58 seconds, the attacker mounted the finance share, copied it in bulk, and unmounted it to cover the trail. All three commands came from PowerShell PID 3728.
| Time | Alert | Command | Network path | Motive |
|---|---|---|---|---|
| 08:40:36 | 1022 | net use Z: \\FILESRV-01\SSF-FinancialRecords |
win-3450 → FILESRV-01 (internal) | Get access to the finance data found during recon |
| 08:41:23 | 1023 | Robocopy.exe . C:\Users\michael.ascot\downloads\exfiltration /E (run from Z:\) |
FILESRV-01 → win-3450 | Stage everything locally, ready to compress and send |
| 08:41:34 | 1024 | net use Z: /delete |
win-3450 | Remove the visible mapped drive so the user or an admin doesn't notice |
Why these are true positives: users don't map drives with PowerShell launched from Downloads, or bulk-copy a finance share into a folder named exfiltration. The working directories tie every command to the phishing victim's Downloads folder.
Escalation: required. Sensitive data was collected from an internal file server as part of an active intrusion.
The 1022 alert landed at Medium severity among dozens of Lows. It was the moment to stop working first-in, first-out and jump to the alert that showed an attack in progress.
DNS exfiltration and data reconstruction (1025–1034)
The attacker zipped the staged files, Base64-encoded the archive, and sent it out about 30 characters at a time as subdomains of haz4rdw4re.io. Rebuilding the 10 nslookup queries recovered exactly what was stolen: two files and a THM{} string.
How DNS exfiltration works: a query like <data>.haz4rdw4re.io is forwarded by the organisation's resolver to the nameserver for haz4rdw4re.io, which the attacker controls, and the attacker reads the data out of the subdomain. DNS is almost always allowed outbound, which is why it works. Motive: get the data past perimeter controls without a direct connection to a suspicious server.
Network path: win-3450 (internal) → internal DNS resolver → haz4rdw4re.io (external, attacker-controlled).
Rebuilding the archive
| Alerts | Decoded content | Meaning |
|---|---|---|
| 1025 | UEsDB… = PK\x03\x04 |
Start of a ZIP archive (first file header) |
| 1026–1028 | ClientPortfolio + Summary.xlsx + data |
File 1: ClientPortfolioSummary.xlsx |
| 1029 | PK\x03\x04, deflate, modified 14 Sep 2023 |
Second file header |
| 1030–1032 | InvestorPresen + tation2023.pptx + data |
File 2: InvestorPresentation2023.pptx |
| 1033–1034 | VEhNe… + …fQ== |
THM{1497321f4f6f059a52dfb124fb16566e} |
Joining chunks 1025–1032 in order and parsing them as a ZIP gives two entries, and both pass CRC verification. That proves the reconstruction is complete and correctly ordered. The file contents are sim placeholders, but the method is the same on a real case.
A quick way to decode the THM{} string:
echo 'VEhNezE0OTczMjFmNGY2ZjA1OWE1MmRmYjEyNGZiMTY1NjZlfQ==' | base64 -d
# THM{1497321f4f6f059a52dfb124fb16566e}
Tip: UEsDB at the start of a Base64 blob means a ZIP file. It's worth learning by sight.
Escalation: required. This is a confirmed data breach of client and investor documents, with likely client-notification and regulatory implications.
The noise: false positives
Thirteen alerts were benign, all from two rules: an email rule that fires on unusual top-level domains, which the SOC lead flagged as untuned, and a parent-child process rule that fires on rare but normal Windows behaviour.
Spam emails (unusual TLD rule)
| Alert | Sender domain | Recipient | Why benign |
|---|---|---|---|
| 1000 | trendymillineryco.me | support@ | Inheritance scam text with no link or attachment; graded as spam, not an incident |
| 1003 | fashionindustrytrends.xyz | yani.zubair | "Grow your hat business" spam, no link |
| 1004 | fashionindustrytrends.xyz | kyra.flores | Time-travel ticket spam, no link |
| 1017 | fashionindustrytrends.xyz | armaan.terry | Same content as 1004, third sender address |
| 1018 | hatventuresworldwide.online | liam.espinoza | "Win a trip" spam, says "click below" but has no link |
The only trigger was the TLD. None of these emails had a payload, so there was no way for them to compromise anyone. The test that separated them from 1005 is payload: an attachment or a link that can run code or harvest credentials.
Windows processes (parent-child rule)
| Alert | Host | Parent → child | What it really is |
|---|---|---|---|
| 1001, 1015 | win-3459, win-3449 | services.exe → TrustedInstaller.exe | Windows Modules Installer (updates) |
| 1002, 1007 | win-3451 | svchost.exe → taskhostw.exe KEYROAMING | Scheduled task for certificate key roaming |
| 1006, 1009 | win-3450, win-3453 | svchost.exe → rdpclip.exe | RDP clipboard monitor, starts with any RDP session |
| 1008 | win-3455 | services.exe → WUDFHost.exe | User-mode driver framework host |
| 1010 | – | – | Benign; misclassified (see below) |
The checks for each: is the binary in its correct path, is the parent the expected one, is the command line clean, and does it spawn anything unusual?
The two misclassifications
- 1010: I assigned myself several alerts at once and didn't realise that one report closes all of them with the same verdict. This benign alert was closed under a true-positive report.
- 1000: I marked the inheritance scam as a true positive because it asked for banking details. The sim graded it as a false positive: in this scenario, payload-free spam from the untuned TLD rule counts as noise, whatever the wording.
Results and feedback
Every true positive was caught and the breach was marked as prevented, finishing 1st with 2735 points (+1730). Two benign alerts marked as true positive pulled the false-positive identification rate down to 85%.
| Metric | Result |
|---|---|
| True positive identification rate | 100% |
| False positive identification rate | 85% |
| Alerts closed | 28 |
| Mean time to resolve | 4 min |
| Mean dwell time | 40 min (longer than average) |
| Slowest alert type | Execution (1024 took 25.85 min) |
The headline "TP rate 47.22%" doesn't match the 100% identification rate shown below it, so it probably measures something else, such as the share of all alerts that were true positives.
AI feedback: the reports were praised as thorough and technically precise, consistently linking each alert back to the phishing email and forward to the exfiltration. Two things to improve:
- Network location: state clearly where activity started and ended from a network point of view (internal host, internal server, external domain). This writeup adds a network path to each stage.
- Attacker motive: state the objective behind each action explicitly rather than leaving it implied. This writeup adds a motive to each stage.
The long dwell time follows from working the queue roughly in order. The Medium execution alerts sat behind a pile of Low noise.
Lessons learned
The biggest lesson: triage the whole queue before working it, because the real attack hid among dozens of Low alerts.
- Prioritise by context, not arrival order. Rank alerts on attack stage (execution and exfiltration above delivery), links to a known compromised user or host, data sensitivity, and how trustworthy the rule is. A Medium execution alert on the phishing victim's host should jump the queue immediately.
- Severity comes from the rule; the verdict comes from the context. PowerView.ps1 (1020) was rated Low, yet it was an attacker tool on a compromised host.
- A delivered payload isn't a closed case. 1005 looked contained when it arrived, but execution showed up 21 minutes later. Revisit delivery alerts once later activity appears for the same user.
- One report closes every alert assigned to you. Only assign several alerts together when they share the same verdict, like the DNS chunks 1025–1034.
- Know your noisy rules. Spam with no payload and normal Windows binaries (TrustedInstaller, taskhostw, rdpclip, WUDFHost) made up most of the queue. Check the path, the parent, the command line and the children, then move on quickly.
- Decode what leaves the network. Rebuilding the DNS chunks turned "data was exfiltrated" into "these two named files were stolen".
UEsDBmeans a ZIP, andVEhNemeansTHM{. - State the where and the why. Every report should name the network path (internal host, internal server, external domain) and the attacker's objective at that step.
About the author
- Software Developer | Learning Cybersecurity | Open for roles * If you're in the early stages of your career in software development (student or still looking for an entry-level role) and in need of mentorship, you can reach out to me.
Comments
Join the discussionNo comments yet. Be the first to comment.



