What is Threat Hunting?
Threat hunting is the proactive, human-led search for attacker activity that has not yet triggered any automated alert. It is the opposite of reactive SOC work — instead of waiting for detections to fire, the threat hunter asks: 'If an adversary were already inside this network, where would they be, and what would they look like?'
Threat hunting is structured thinking about attacker behaviour, expressed as data queries. The best hunters are curious, methodical, and deeply familiar with both attacker TTPs and their own environment's normal.
Most mature SOC programmes run three tiers: L1/L2 analysts handle alert triage, incident responders investigate confirmed events, and threat hunters proactively search in the gap between alerts. The gap is where advanced persistent threats live for months.
The Hunting Methodology
Every hunt starts with a hypothesis — a specific, testable statement about attacker behaviour. Without a hypothesis, you are not hunting; you are browsing. Good hypotheses are derived from threat intelligence, ATT&CK techniques, or anomalies noticed during routine work.
Start from a specific ATT&CK technique or threat intel report. E.g., 'APT29 uses scheduled tasks for persistence — are there any unusual schtasks on endpoints?'
Compare current behaviour to historical norms. Unusually high DNS query volumes, new parent-child process relationships, or first-time-seen executables.
Receive an IOC from a threat intel feed or ISAC. Hunt for that specific indicator across your environment before it fires an alert.
After a major vulnerability disclosure or geopolitical event, proactively hunt for exploitation patterns targeting your industry.
Using ATT&CK Effectively
MITRE ATT&CK is the threat hunter's atlas. It catalogues 14 tactic categories — from Initial Access through Impact — each containing techniques and sub-techniques used by real adversaries. The key is using it as a prioritisation framework, not reading it cover-to-cover.
| Tactic | ID | Hunt Priority | Key Techniques to Hunt |
|---|---|---|---|
| Initial Access | TA0001 | High | Phishing (T1566), External Remote Services (T1133) |
| Execution | TA0002 | Critical | PowerShell (T1059.001), WMI (T1047), Scheduled Tasks (T1053) |
| Persistence | TA0003 | Critical | Registry Run Keys (T1547.001), New Services (T1543) |
| Credential Access | TA0006 | Critical | LSASS Dump (T1003.001), Kerberoasting (T1558.003) |
| Lateral Movement | TA0008 | High | Pass-the-Hash (T1550.002), PsExec (T1021.002) |
| Exfiltration | TA0010 | High | Web Service (T1567), DNS Tunnelling (T1048.003) |
Use the ATT&CK Navigator to build a heatmap of your current detection coverage. Any technique with no coverage is a potential hunting hypothesis.
Writing Good Hypotheses
The quality of your hunt is bounded by the quality of your hypothesis. A good hypothesis is specific, falsifiable, and tied to real attacker behaviour.
| Bad Hypothesis | Good Hypothesis | Why Better |
|---|---|---|
| 'Attackers might be in our network' | 'APT41 uses PowerShell -EncodedCommand for execution — are there encoded PS commands executed outside business hours on endpoints?' | Specific technique, timeframe, and asset scope |
| 'Check DNS for weird stuff' | 'DNS queries with payloads >100 bytes in the subdomain field may indicate DNS tunnelling — find hosts with unusually long average query lengths' | Measurable threshold, specific IOC pattern |
| 'Look for malware' | 'LockBit uses vssadmin to delete shadow copies before encryption — are there any vssadmin calls in the last 7 days not from a backup job?' | Specific tool, specific action, known legitimate baseline |
Detection Queries — Splunk SPL
Once the hypothesis is written, translate it into queries against your data sources. Here are three production-ready hunting queries:
| Hunt: T1059.001 — PowerShell EncodedCommand
index=windows EventCode=4688
CommandLine IN ("*-EncodedCommand*", "*-enc *", "*-e *")
NOT user IN ("NT AUTHORITY\\SYSTEM", "svc_backup")
| eval hour=strftime(_time, "%H")
| where hour < 7 OR hour > 20 | Outside business hours
| stats count by ComputerName, user, CommandLine
| sort -count| Hunt: T1053.005 — Suspicious Scheduled Task
index=windows EventCode=4698
| rex field=TaskContent "(?P[^<]+) "
| where match(cmd, "(?i)(powershell|cmd|wscript|cscript|mshta|regsvr32)")
| table _time, ComputerName, SubjectUserName, TaskName, cmd
| sort -_time| Hunt: T1021.002 — SMB Lateral Movement
index=windows EventCode=7045
ServiceName IN ("PSEXESVC", "paexec", "remcom")
| join ComputerName
[search index=windows EventCode=4624 LogonType=3]
| table _time, ComputerName, AccountName, ServiceName, ImagePath
| sort -_timeMeasuring Hunt Effectiveness
| Metric | Description | Target |
|---|---|---|
| Hypotheses per month | Number of distinct hunts conducted | ≥4 per hunter per month |
| Confirmed findings rate | Hunts that found real attacker activity or unknown risk | ≥20% of hunts |
| New detections created | Sigma/SPL rules written from hunt findings | ≥1 per 3 hunts |
| ATT&CK coverage increase | Technique coverage % growth quarter-over-quarter | +5% per quarter |
| Dwell time reduction | Average time attackers present before detection | Target: <24h |
A hunt that finds nothing is not a failed hunt. Absence of evidence builds confidence in your detection coverage and documents the work for audit purposes.
Getting Started Today
You do not need a dedicated threat hunting team to start. Any L2 analyst with SIEM access can run their first hunt in an afternoon.
- Pick one ATT&CK technique relevant to your industry threat profile
- Write a one-sentence hypothesis — 'If technique X is being used, I expect to see Y in data source Z'
- Run the query — investigate anything anomalous
- Document the hunt — findings or no findings, write it up
- Convert findings to detections — every successful hunt should produce a permanent detection rule
Start with T1059 (Scripting/Command Execution) — it has excellent data availability in Windows Event Logs and is used in almost every intrusion. Your first hunt will teach you more than 10 hours of reading.