πΉ Proactive Adversary Pursuit, Behavioral Outlier Stacking & Empirical Hypothesis Testing
Traditional Security Operations Center (SOC) workflows rely primarily on reactive detection: alerting rules, signature patterns, and correlation thresholds waiting for an adversary to trip an automated alarm. However, advanced adversaries, persistent threat actors, and insider threats deliberately operate below these static thresholdsβleveraging Living-off-the-Land Binaries (LOLBins), unbacked in-memory process manipulation, stolen administrative credentials, and valid administrative scripting utilities (PowerShell, WMI) that blend seamlessly into normal enterprise baseline noise. Cyber Threat Hunting is the iterative, human-led, and hypothesis-driven process of proactively querying security telemetry to detect malicious activity that has evaded existing automated defenses. Rather than waiting for alerts, threat hunters operate under the assumption of breach: formulating testable propositions derived from Cyber Threat Intelligence (CTI), executing analytical methodologies such as Long-Tail Baseline Stacking, navigating standardized models like the TaHiTI Framework, and codifying confirmed discoveries directly into durable automated SIEM/EDR detections.
SECURITY OPERATIONS PARADIGM COMPARISON:
βββββββββββββββββββββββ¬βββββββββββββββββββββββββββ¬βββββββββββββββββββββββ¬ββββββββββββββββββββββββββββββββββ
β OPERATIONAL VECTOR β TRADITIONAL SOC (TIER 1) β INCIDENT RESPONSE β PROACTIVE THREAT HUNTING β
βββββββββββββββββββββββΌβββββββββββββββββββββββββββΌβββββββββββββββββββββββΌββββββββββββββββββββββββββββββββββ€
β Triggering Event β Reactive: Automated SIEM β Reactive: Escalated β Proactive: Empirical β
β β or EDR alert threshold β confirmed incident β hypothesis formulated by team β
βββββββββββββββββββββββΌβββββββββββββββββββββββββββΌβββββββββββββββββββββββΌββββββββββββββββββββββββββββββββββ€
β Operational Mindset β "Did the tool detect an β "The system is hit; β "The adversary is ALREADY in β
β β anomalous signature?" β stop active spread!" β the network; find the trace!" β
βββββββββββββββββββββββΌβββββββββββββββββββββββββββΌβββββββββββββββββββββββΌββββββββββββββββββββββββββββββββββ€
β Primary Data Source β Normalized alert queues, β Forensic disk/memory β Raw telemetry lakes: Sysmon, β
β β static rule outputs β images, jump bags β NetFlow, process execution logs β
βββββββββββββββββββββββΌβββββββββββββββββββββββββββΌβββββββββββββββββββββββΌββββββββββββββββββββββββββββββββββ€
β Methodology β Playbook-driven triage, β Contain, Eradicate, β Outlier stacking, baseline β
β β alert dismissal / escal. β Recover (NIST/SANS) β deviations, behavioral sweeps β
βββββββββββββββββββββββΌβββββββββββββββββββββββββββΌβββββββββββββββββββββββΌββββββββββββββββββββββββββββββββββ€
β Final Deliverable β Closed alert ticket, β Restored services, β New automated detection rule, β
β β false-positive exclusion β formal RCA report β uncovered adversary, gap patch β
βββββββββββββββββββββββ΄βββββββββββββββββββββββββββ΄βββββββββββββββββββββββ΄ββββββββββββββββββββββββββββββββββ
THE SQRRL HUNTING MATURITY MODEL (HMM) ARCHITECTURE:
LEVEL 4: LEADING (FULLY AUTOMATED & SYSTEMATIC)
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β β’ High automation: Feedback loops automatically operationalize successful hunts into detections β
β β’ Custom analytics: Data science, statistical clustering, and continuous baseline modeling. β
β β’ Dedicated hunt team: Constantly executing structured hypothesis campaigns across enterprise. β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β²
LEVEL 3: INNOVATIVE (METHODICAL & DATA-DRIVEN) β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β β’ Structured processes: Formalized frameworks (TaHiTI, MITRE ATT&CK). β
β β’ Custom data extraction: Analysts write custom queries across massive centralized data lakes.β
β β’ High data collection: Endpoint process tracking (Sysmon), cloud trails, deep network flows. β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β²
LEVEL 2: PROCEDURAL (FOLLOWING PLAYBOOKS) β
ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β β’ Follows established procedures published by third parties (e.g., executing public hunting guides). β
β β’ Moderate data collection: Centralized SIEM with standard OS logs and proxy traffic. β
β β’ Low-to-moderate analytical autonomy: Can repeat known hunts, but limited novel hypothesis gen. β
ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β²
LEVEL 1: MINIMAL (SEARCHING ON ATOMIC IOCs) β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β β’ Search capability: Queries SIEM for atomic indicators (IPs, hashes) shared in threat reportsβ
β β’ Low data collection: Limited endpoint visibility; predominantly perimeter firewall logs. β
β β’ Purely indicator-driven: Little to no behavioral or TTP-based investigative depth. β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β²
LEVEL 0: INITIAL (REACTIVE ONLY - NO HUNTING) β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β β’ 100% Reactive: Relies completely on automated alerts from Antivirus, EDR, and SIEM rules. β
β β’ Zero threat hunting capability: No hypothesis generation; no proactive queries performed. β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
THE TaHiTI THREAT HUNTING LIFECYCLE (PHASE SPECIFICATION):
PHASE 1: TRIGGER & INITIATION
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β β’ Ingestion of Triggers: Threat Intelligence reports, MITRE ATT&CK TTP updates, newly β
β discovered red team vectors, industry incident disclosures, or internal architectural changes β
β β’ Scoping & Prioritization: Evaluate feasibility, data availability, and organizational risk. β
β β’ Hypothesis Formulation: State the testable assumption clearly: β
β "An adversary is executing T1059.001 (PowerShell) using Base64 encoded flags to bypass CLI β
β inspection and establish C2 beacons on domain controller systems." β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β
βΌ
PHASE 2: HUNT EXECUTION (THE ANALYTICAL CORE)
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β β’ Data Requirement Validation: Verify necessary telemetry exists (Sysmon Event ID 1 / 4688). β
β β’ Query Formulation: Craft high-precision analytical searches (Splunk SPL, Elastic EQL). β
β β’ Execute Analytical Techniques: β
β βββ Outlier / Long-Tail Analysis : Stack rare process names and command-line arguments. β
β βββ Baseline Deviation : Identify unusual parent-child execution pairings. β
β βββ Temporal Clustering : Isolate beaconing intervals and persistence cadences. β
β β’ Evidence Evaluation: Distinguish verified malicious actions from administrative anomalies. β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β
βΌ
PHASE 3: OPERATIONALIZATION & FINALIZATION
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β A. IF THREAT DISCOVERED (TRUE POSITIVE): β
β β’ Immediately escalate to Incident Response team with exact host and process artifacts. β
β B. OPERATIONAL HARDENING (EVERY HUNT MUST PRODUCE AN ASSET): β
β β’ Codify discovery logic into a durable Sigma detection rule for continuous SIEM alerting. β
β β’ Document telemetry blind spots (e.g., "Script block logging disabled on Subnet B"). β
β β’ Publish formal Hunt Report documenting hypothesis, query syntax, findings, and metrics. β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
THE THREE PILLARS OF HYPOTHESIS FORMULATION:
1. INTELLIGENCE-DRIVEN HYPOTHESIS:
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β β’ Source: Cyber Threat Intelligence reporting on active campaigns (e.g., FIN7, APT29). β
β β’ Context: CTI reports FIN7 is targeting financial endpoints using weaponized LNK shortcut β
β files dropping malicious DLLs via DLL Side-Loading (T1574.002). β
β β’ FORMAL HYPOTHESIS: β
β "Adversaries are leveraging legitimate, signed Windows binaries (e.g. spoolsv.exe) located β
β in non-standard directories to load unsigned rogue DLLs from user-writable directories." β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
2. THREAT-MODEL-DRIVEN HYPOTHESIS:
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β β’ Source: Internal risk assessments and crown-jewel architecture reviews. β
β β’ Context: Active Directory Domain Controllers host critical NTDS.dit credential databases. β
β β’ FORMAL HYPOTHESIS: β
β "An insider threat or compromised service account is attempting to extract domain password β
β hashes via Volume Shadow Copies (vssadmin / ntdsutil) outside of scheduled maintenance." β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
3. SITUATIONAL / ANOMALY-DRIVEN HYPOTHESIS:
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β β’ Source: Environmental posture shifts, new software rollouts, or unusual network traffic. β
β β’ Context: Organizations transitioning to remote desktop gateways. β
β β’ FORMAL HYPOTHESIS: β
β "Adversaries are executing internal lateral movement by tunneling RDP traffic through non- β
β standard outbound ports (e.g. TCP 443, 8080) using living-off-the-land proxy utilities." β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
THE FALSIFIABILITY CHECKLIST FOR HUNT TEAMS:
1. Is the hypothesis specific? (Identifies explicit TTPs, not just "something bad is happening").
2. Is the required telemetry collected? (Auditd, Sysmon, Zeek logs available and retention > 30d).
3. Can the hypothesis be proven FALSE? (Defines a clear boundary separating benign baseline from threat).
LONG-TAIL ANALYSIS (BASELINE STACKING) ARCHITECTURE:
THE CORE PREMISE OF STACKING:
In an enterprise of 10,000 workstations, legitimate business applications execute thousands
of times daily across all systems (The "Head" of the distribution).
Malicious implants, backdoors, and adversary scripts execute on only ONE or TWO machines
(The "Long-Tail" of the distribution).
FREQUENCY DISTRIBUTION PLOT:
Event
Count
β²
β β
β β Legitimate Enterprise Software
β β (head of distribution: svchost, chrome, outlook, teams)
β β Count: 50,000+ executions
β β
β β
β β
β β THE LONG-TAIL (HUNT TARGET)
β β Rare executions: Count = 1 or 2
β β β β β
β β β β β β βΌ
βββ΄βββ΄βββ΄βββ΄βββ΄βββ΄βββ΄βββ΄βββ΄βββ΄βββ΄βββ΄βββ΄βββ΄βββ΄βββ΄βββ΄βββ΄βββ΄βββΊ Distinct Process Command-Lines
SAMPLE ENTERPRISE STACKING AUDIT (PowerShell Executions Grouped by Command-Line):
βββββββββ¬ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β COUNT β PROCESS COMMAND-LINE EXECUTION STRING β
βββββββββΌββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ€
β 14200 β powershell.exe -ExecutionPolicy Restricted -Command Get-Date β
β 8900 β powershell.exe -NoProfile -NonInteractive -File C:\Program Files\SCCM\client.ps1 β
β 3100 β powershell.exe -Command "& {Get-WmiObject -Class Win32_Processor}" β
βββββββββΌββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ€
β 1 β powershell.exe -nop -w hidden -enc aW52b2tlLW1pbWlrYXR6... β
β β (EXECUTED ONLY ONCE ON ENTIRE NETWORK! IMMEDIATE HIGH-CONFIDENCE MALICIOUS ARTIFACT!) β
βββββββββ΄ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
word.exe spawning powershell.exe) using Windows Security Event ID 4688 and Sysmon Event IDs 1, 3, 7, and 13.
HUNTING ANOMALOUS PROCESS PARENTAGE IN WINDOWS TELEMETRY:
LEGITIMATE WINDOWS PROCESS PARENTAGE BASELINES:
βββββββββββββββββββββββββββ¬βββββββββββββββββββββββββββββββ¬βββββββββββββββββββββββββββββββββββββββ
β PROCESS NAME β EXPECTED PARENT PROCESS β ABERRANT / MALICIOUS PARENT SPOOFING β
βββββββββββββββββββββββββββΌβββββββββββββββββββββββββββββββΌβββββββββββββββββββββββββββββββββββββββ€
β smss.exe β System (PID 4) β Spawed by user shell (cmd.exe) β
β csrss.exe β smss.exe β Spawned by explorer.exe β
β wininit.exe β smss.exe β Spawned by svchost.exe β
β services.exe β wininit.exe β Spawned by powershell.exe β
β lsass.exe β wininit.exe β Spawned by any user utility / malwareβ
β explorer.exe β userinit.exe β Spawned by cmd.exe / wscript.exe β
β svchost.exe β services.exe β Spawned by winword.exe / excel.exe β
βββββββββββββββββββββββββββ΄βββββββββββββββββββββββββββββββ΄βββββββββββββββββββββββββββββββββββββββ
HUNTING QUERY 1: OFFICE APPLICATION SPAWNING SCRIPT ENGINES (T1059 / T1204)
Splunk SPL:
index=endpoint EventCode=1 (ParentImage="*\\winword.exe" OR ParentImage="*\\excel.exe")
Image IN ("*\\powershell.exe", "*\\cmd.exe", "*\\wscript.exe", "*\\cscript.exe", "*\\mshta.exe")
| stats count by host, user, ParentImage, Image, CommandLine
HUNTING QUERY 2: UNUSUAL ACCESS TO LSASS MEMORY (SYSMON EVENT ID 10)
Sysmon Event ID 10 logs when one process opens another process handle.
Elastic EQL:
process where event.code == "10" and
winlog.event_data.TargetImage endswith "lsass.exe" and
not winlog.event_data.SourceImage in (
"?:\\Windows\\System32\\svchost.exe",
"?:\\Program Files\\Windows Defender\\MsMpEng.exe"
) and
/* Catch read/query permissions used by credential dumping tools */
(winlog.event_data.GrantedAccess == "0x1010" or winlog.event_data.GrantedAccess == "0x1fffff")
sudo, and hunting web shells spawned by web server daemons (www-data).
LINUX HUNTING TELEMETRY & AUDITD CONFIGURATION (/etc/audit/rules.d/audit.rules):
1. CRITICAL AUDITD HUNTING PROBES:
# Monitor execution of commands by unprivileged users attempting privilege escalation
-a always,exit -F arch=b64 -S execve -F euid=0 -F auid>=1000 -k root_escalation
# Monitor modifications to critical security and authentication files
-w /etc/passwd -p wa -k identity_modification
-w /etc/shadow -p wa -k identity_modification
-w /etc/sudoers -p wa -k sudoers_modification
-w /root/.ssh/authorized_keys -p wa -k ssh_key_persistence
2. HUNTING WEB SHELLS (WEB SERVER DAEMON SPAWNING SHELLS - T1505.003):
Threat Context: Adversary uploads PHP web shell and runs interactive commands:
Web Daemon User : www-data, apache, or nginx
Anomalous Child : /bin/sh, /bin/bash, python, curl, wget, whoami
Auditd Hunting Query (ausearch):
$ sudo ausearch -k web_activity -m EXECVE | grep -E "uid=33\(www-data\).*comm=\"(bash|sh|whoami|curl)\""
Normalized SIEM Hunting Query (Splunk SPL):
index=linux_auditd (user="www-data" OR user="nginx" OR user="apache")
exe IN ("/bin/bash", "/bin/sh", "/usr/bin/python*", "/usr/bin/whoami", "/usr/bin/curl")
| stats count, values(proctitle) as executed_commands by host, ppid, exe, user
| where count < 10
import collections
from typing import List, Dict, Any
class ThreatHuntStacker:
"""
Automated Long-Tail Threat Hunting Analytics Engine:
1. Aggregates enterprise process execution telemetry across fleet.
2. Groups and counts exact command-line executions.
3. Calculates frequency percentage across total dataset.
4. Flags statistical outliers executing on <= rarity_threshold systems.
"""
def __init__(self, rarity_threshold: int = 1):
self.rarity_threshold = rarity_threshold
def execute_long_tail_stack(self, execution_logs: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
total_events = len(execution_logs)
# Counter maps (process_name, command_line) -> frequency
command_counts = collections.Counter()
host_tracking = collections.defaultdict(set)
for event in execution_logs:
key = (event["process_name"], event["command_line"])
command_counts[key] += 1
host_tracking[key].add(event["host"])
outliers = []
for (proc, cmd), count in command_counts.items():
if count <= self.rarity_threshold:
prevalence_pct = (count / total_events) * 100
outliers.append({
"process": proc,
"command_line": cmd,
"frequency_count": count,
"prevalence_percentage": round(prevalence_pct, 4),
"affected_hosts": list(host_tracking[(proc, cmd)])
})
# Sort outliers ascending (rarest events first)
outliers.sort(key=lambda x: x["frequency_count"])
return outliers
if __name__ == "__main__":
print("==========================================================================")
print(" THREAT HUNTING: AUTOMATED LONG-TAIL STACKING ENGINE")
print("==========================================================================")
# Simulated enterprise process dataset (thousands of events modeled)
mock_telemetry = [
{"host": "WS-01", "process_name": "svchost.exe", "command_line": "svchost.exe -k DcomLaunch"},
{"host": "WS-02", "process_name": "svchost.exe", "command_line": "svchost.exe -k DcomLaunch"},
{"host": "WS-03", "process_name": "svchost.exe", "command_line": "svchost.exe -k DcomLaunch"},
{"host": "WS-04", "process_name": "chrome.exe", "command_line": "chrome.exe --type=utility"},
{"host": "WS-05", "process_name": "chrome.exe", "command_line": "chrome.exe --type=utility"},
{"host": "WS-01", "process_name": "powershell.exe", "command_line": "powershell.exe -w hidden -enc JABzAD0ATg..."},
{"host": "WS-02", "process_name": "whoami.exe", "command_line": "whoami /priv"}
]
stacker = ThreatHuntStacker(rarity_threshold=1)
results = stacker.execute_long_tail_stack(mock_telemetry)
print(f"[*] Ingested {len(mock_telemetry)} events. Analyzing statistical distribution...")
print(f"[!] Discovered {len(results)} statistical outlier(s) (Frequency <= 1 execution):\n")
for idx, outlier in enumerate(results, 1):
print(f"Outlier #{idx}:")
print(f" βββ Process : {outlier['process']}")
print(f" βββ Command Line : {outlier['command_line']}")
print(f" βββ Count : {outlier['frequency_count']} execution(s) across fleet ({outlier['prevalence_percentage']}%)")
print(f" βββ Host(s) : {outlier['affected_hosts']}")
THREAT HUNTING ANTI-PATTERNS & TRIAGE PROTOCOLS:
// ==============================================================================
// PITFALL 1: HUNTING WITHOUT A TESTABLE HYPOTHESIS (THE "DATA FISHING" TRAP)
// MISTAKE: Analyst opens SIEM and randomly types: "cmd.exe" or "error".
// CONSEQUENCE: Returns 400,000 unorganized events. Analyst browses randomly for 3 hours,
// finds nothing conclusive, and wastes operational capacity without documenting findings.
// REMEDIATION: Mandate Formal Hypothesis Formulation:
// Every hunt must declare a documented hypothesis mapped to MITRE ATT&CK before running queries:
// β’ Hypothesis : "Adversaries are establishing persistence via Registry Run Keys (T1547.001)"
// β’ Scope : Workstations in Engineering OU over the last 14 days
// β’ Telemetry : Sysmon Event ID 13 (RegistryEvent: ValueSet)
// ==============================================================================
// ==============================================================================
// PITFALL 2: IGNORING ADMINISTRATIVE LIVING-OFF-THE-LAND UTILITIES (NOISE FATIGUE)
// MISTAKE: Threat hunter queries for certutil.exe and finds 500 executions.
// Assumes the network is completely compromised and escalates a false crisis!
// CAUSE: SCCM, IT update scripts, or deployment tools frequently use certutil.exe
// to download internal SSL certificates.
// REMEDIATION: Contextual Baseline Whitelisting:
// Filter out authorized IT execution paths (signed binaries, verified script hashes,
// authorized parent PIDs like sccm_exec.exe) to isolate anomalous command-line arguments.
// ==============================================================================
// ==============================================================================
// PITFALL 3: THE "DISCOVER AND FORGET" TRAP (FAILING TO OPERATIONALIZE)
// MISTAKE: A threat hunter spends 2 weeks uncovering a sophisticated WMI persistence
// mechanism. They manually remove the registry key on the single host and close the file.
// CATASTROPHE: 6 months later, another attacker uses the exact same WMI technique, and
// the SOC remains completely blind!
// REMEDIATION: Mandatory Operationalization Milestone:
// Every successful hunt MUST convert its detection logic into a durable automated asset:
// 1. Author a Sigma detection rule for the SIEM.
// 2. Add behavioral telemetry alerts into the EDR console.
// 3. Document the technique in the corporate Threat Hunting Knowledge Base.
// ==============================================================================
THREAT HUNTING FRAMEWORKS & METHODOLOGY MATRIX:
βββββββββββββββββββββββββββ¬βββββββββββββββββββββββββββββββ¬ββββββββββββββββββββββ¬ββββββββββββββββββββββββββββββββ
β FRAMEWORK / MODEL β LEVEL / CATEGORY β TARGET OBJECTIVE β CORE OPERATIONAL FOCUS β
βββββββββββββββββββββββββββΌβββββββββββββββββββββββββββββββΌββββββββββββββββββββββΌββββββββββββββββββββββββββββββββ€
β Sqrrl Hunting Maturity β HMM Levels 0 to 4 β Organizational β Evaluates hunt posture from β
β Model (HMM) β Initial to Leading β Maturity Bench. β reactive alerting to automatedβ
βββββββββββββββββββββββββββΌβββββββββββββββββββββββββββββββΌββββββββββββββββββββββΌββββββββββββββββββββββββββββββββ€
β TaHiTI Framework β Trigger β Execution β β Operational β Structured threat hunting β
β (Dutch Banking Sector) β Operationalization β Lifecycle β project management methodologyβ
βββββββββββββββββββββββββββΌβββββββββββββββββββββββββββββββΌββββββββββββββββββββββΌββββββββββββββββββββββββββββββββ€
β Long-Tail Analysis β Stacking Technique β Analytical β Aggregates fleet telemetry to β
β (Baseline Stacking) β Rare Execution Outliers β Anomaly Hunt β isolate rare (count <= 2) runsβ
βββββββββββββββββββββββββββΌβββββββββββββββββββββββββββββββΌββββββββββββββββββββββΌββββββββββββββββββββββββββββββββ€
β MITRE ATT&CK β Enterprise Matrix β Behavioral β Provides standardized TTP β
β Mapping β Tactics & Techniques (T1059) β Taxonomy Anchor β anchors for testable hypothesesβ
βββββββββββββββββββββββββββΌβββββββββββββββββββββββββββββββΌββββββββββββββββββββββΌββββββββββββββββββββββββββββββββ€
β Windows Sysmon β Event IDs 1, 3, 7, 8, 10, 11 β Endpoint Hunting β Deep process lineage, remote β
β Telemetry β Process creation & injection β Telemetry Data β thread injection, memory readsβ
βββββββββββββββββββββββββββΌβββββββββββββββββββββββββββββββΌββββββββββββββββββββββΌββββββββββββββββββββββββββββββββ€
β Linux Auditd / eBPF β syscall execve, openat β Linux Infrastructureβ Tracks root privilege shifts, β
β Telemetry β Proctitle monitoring β Telemetry Data β web server shell spawning β
βββββββββββββββββββββββββββΌβββββββββββββββββββββββββββββββΌββββββββββββββββββββββΌββββββββββββββββββββββββββββββββ€
β Detection-as-Code β Sigma / YARA Rules β Post-Hunt β Converts human hunt findings β
β Operationalization β Git-versioned detection rulesβ Automation Asset β into permanent SIEM detectionsβ
βββββββββββββββββββββββββββ΄βββββββββββββββββββββββββββββββ΄ββββββββββββββββββββββ΄ββββββββββββββββββββββββββββββββ€
β TOP 5 THREAT HUNTING COMMANDMENTS: β
β 1. Assume Active Breach: Formulate hypotheses based on the premise that defenses were already bypassed. β
β 2. Anchor in MITRE ATT&CK: Base testable propositions on real-world adversary behaviors, not hashes. β
β 3. Stack the Long-Tail: Hunt for anomalies by filtering out the high-frequency baseline head. β
β 4. Track Process Lineage: Inspect anomalous parent-child relationships (e.g. Word spawning shells). β
β 5. Always Operationalize: Formalize every confirmed finding into an automated SIEM detection rule. β
ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ