Alert triage in Elastic Security is the systematic process of reviewing, classifying, and prioritizing security alerts to determine which represent genuine threats. Elastic's AI-driven Attack Discovery feature can triage hundreds of alerts down to discrete attack chains, but skilled analyst triage remains essential. A structured triage workflow typically takes 5-10 minutes per alert cluster using Elastic's built-in tools.
When to Use
When conducting security assessments that involve performing alert triage with elastic siem
When following incident response procedures for related security events
When performing scheduled security testing or auditing activities
When validating security controls through hands-on testing
Detection Gaps & Validation
ECS field gaps break the rule, not just the view: if a source isn't normalized to ECS, fields like user.name, process.parent.name, source.ip, or host.name are null and the alert renders with empty context — or the detection rule never matches. Confirm the index has the ECS fields the rule keys on (run the ES|QL KEEP/STATS and check for nulls) before trusting a "no related activity" result.
Benign true positive ≠ false positive: an admin running PowerShell or a vuln scanner generating auth failures is expected behavior; closing it as FP creates a tuning task that blinds you to the real thing. Classify it benign-TP and exception the asset/user, don't broaden the rule.
Single-alert tunnel vision: triagers judge one alert and miss that Attack Discovery / .alerts-security.alerts-default shows it as one node in a chain (failed logins → success → lateral movement). Always pivot on source.ip/user.name over 24h before classifying.
Risk score is not severity:kibana.alert.risk_score is rule-authored and uncalibrated to your asset criticality — a 73 on a domain controller outranks a 90 on a test box. Cross-reference asset context, don't triage by score alone.
Validate the verdict before escalating: confirm the matched indicator is current (TI logs-ti_* not stale), check event.outcome is actually success, and confirm the process tree / parent process supports the malicious narrative rather than a clipped single event.
Prerequisites
Elastic Security deployed (version 8.x or later)
Elastic Agent or Beats configured for endpoint and network data collection
Detection rules enabled and generating alerts
Elastic Common Schema (ECS) compliance across data sources
Analyst access to Kibana Security app with appropriate privileges
Alert Triage Workflow
Step 1: Initial Alert Assessment (2 minutes)
When viewing an alert in Elastic Security, review the alert details panel:
Alert Details Panel:
- Rule Name and Description
- Severity and Risk Score
- MITRE ATT&CK Mapping
- Host and User Context
- Process Tree (for endpoint alerts)
- Timeline of related events
Key Fields to Examine First
Field
Purpose
ECS Field
Rule severity
Initial priority assessment
kibana.alert.severity
Risk score
Quantified threat level
kibana.alert.risk_score
Host name
Affected system
host.name
User name
Affected identity
user.name
Process name
Executing process
process.name
Source IP
Origin of activity
source.ip
Destination IP
Target of activity
destination.ip
MITRE tactic
Attack stage
threat.tactic.name
Step 2: Context Gathering (3 minutes)
Query Related Events with ES|QL
FROM logs-endpoint.events.*
| WHERE host.name == "affected-host" AND @timestamp > NOW() - 1 HOUR
| STATS count = COUNT(*) BY event.category, event.action
| SORT count DESC
Find All Activity from Suspicious User
FROM logs-*
| WHERE user.name == "suspicious-user" AND @timestamp > NOW() - 24 HOURS
| STATS count = COUNT(*), unique_hosts = COUNT_DISTINCT(host.name) BY event.category
| SORT count DESC
Check for Related Alerts from Same Source
FROM .alerts-security.alerts-default
| WHERE source.ip == "10.0.0.50" AND @timestamp > NOW() - 24 HOURS
| STATS alert_count = COUNT(*) BY kibana.alert.rule.name, kibana.alert.severity
| SORT alert_count DESC
Investigate Lateral Movement from Same IP
FROM logs-system.auth-*
| WHERE source.ip == "10.0.0.50" AND event.outcome == "success"
| STATS login_count = COUNT(*), hosts = COUNT_DISTINCT(host.name) BY user.name
| WHERE hosts > 3
FROM logs-ti_*
| WHERE threat.indicator.ip == "203.0.113.50"
| KEEP threat.indicator.type, threat.indicator.provider, threat.indicator.confidence, threat.feed.name
Check File Hash Against Known Threats
FROM logs-endpoint.events.file-*
| WHERE file.hash.sha256 == "abc123..."
| STATS occurrences = COUNT(*) BY host.name, file.path, user.name
Step 4: Classification Decision (2 minutes)
Classification
Criteria
Action
True Positive
Confirmed malicious activity
Escalate to incident, begin containment
Benign True Positive
Expected behavior matching rule
Document in alert notes, acknowledge
False Positive
Rule triggered on benign activity
Mark as false positive, create tuning task
Needs Investigation
Insufficient data for determination
Assign for deeper investigation
Step 5: Documentation and Escalation (1 minute)
For each triaged alert, document:
Classification decision with rationale
Evidence artifacts examined
Related alerts or investigations
Recommended next steps
Detection Rules for Triage
Pre-Built Detection Rules
Elastic Security includes 1000+ pre-built detection rules organized by:
MITRE ATT&CK Tactic: Initial Access, Execution, Persistence, etc.
Platform: Windows, Linux, macOS, Cloud
Data Source: Endpoint, Network, Cloud, Identity
Custom Alert Correlation Rule
{"name":"Multiple Failed Logins Followed by Success","type":"threshold","query":"event.category:authentication AND event.outcome:failure","threshold":{"field":["source.ip","user.name"],"value":5,"cardinality":[{"field":"user.name","value":3}]},"severity":"high","risk_score":73,"threat":[{"framework":"MITRE ATT&CK","tactic":{"id":"TA0006","name":"Credential Access"},"technique":[{"id":"T1110","name":"Brute Force"}]}]}
AI-Assisted Triage
Elastic AI Assistant Integration
Open alert in Elastic Security
Click AI Assistant panel
Use quick prompts:
"Summarize this alert" - Get initial assessment
"Generate ES|QL query to find related activity" - Expand investigation
"What are the recommended response actions?" - Get playbook guidance
"Is this likely a false positive?" - Get AI confidence assessment
