Correlated Alerts on Similar User Identities

IMPORTANT: This documentation is no longer updated. Refer to Elastic's version policy and the latest documentation.

Correlated Alerts on Similar User Identities

This rule correlates alerts from multiple integrations and event categories that involve different user.name values which may represent the same real-world identity. It uses an LLM-based similarity analysis to evaluate whether multiple user identifiers (e.g. naming variations, formats, aliases, or domain differences) likely belong to the same person.

Rule type: esql

Rule indices: None

Severity: high

Risk score: 73

Runs every: 30m

Searches indices from: now-60m (Date Math format, see also Additional look-back time)

Maximum alerts per execution: 100

References:

Tags:

Domain: Identity
Domain: LLM
Use Case: Threat Detection
Use Case: Identity and Access Audit
Resources: Investigation Guide
Rule Type: Higher-Order Rule

Version: 2

Rule authors:

Elastic

Rule license: Elastic License v2

Investigation guide

edit

Triage and analysis

Disclaimer: This investigation guide was created using generative AI technology and has been reviewed to improve its accuracy and relevance. While every effort has been made to ensure its quality, analysts should validate findings against their environment and identity architecture.

Investigating Correlated Alerts on Similar User Identities

This rule identifies alerts from multiple integrations and event categories involving different user.name values that may represent the same real-world identity. An LLM is used to assess string similarity and naming patterns to determine whether multiple user identifiers likely belong to the same person, which may indicate account compromise, credential abuse, or identity misuse across systems.

Possible investigation steps

Review the correlated user.name values and validate whether they represent naming variations, aliases, or identity mappings.
Examine the LLM output fields (verdict, confidence, summary) as decision support, not ground truth.
Analyze the diversity of alert sources, event categories, and detection rules involved.
Reconstruct the alert timeline to identify potential stages such as initial access, lateral movement, privilege escalation, or persistence.
Correlate with authentication logs, IAM/SSO telemetry, EDR data, and network logs to identify shared sessions, IPs, devices, or hosts.
Validate identities against directory services, identity providers, and federation mappings.

False positive analysis

Identity format variations across systems (e.g., first.last, flast, user@domain).
Federated identity mappings between on-prem, cloud, and SaaS platforms.
Service, automation, and CI/CD accounts with similar naming conventions.
Separate admin and standard user accounts for the same individual.
Shared credentials or naming templates in development and test environments.

Response and remediation

Temporarily disable or suspend correlated accounts if compromise is suspected.
Revoke active sessions, tokens, and credentials.
Investigate access scope, privileges, and lateral movement paths.
Perform endpoint and identity forensics to identify persistence mechanisms.
Remediate IAM misconfigurations and federation issues.
Enhance monitoring for identity correlation, credential misuse, and cross-platform abuse..

Setup

edit

Setup

LLM Configuration

This rule uses the ES|QL COMPLETION command with Elastic’s managed General Purpose LLM v2 (.gp-llm-v2-completion), which is available out-of-the-box in Elastic Cloud deployments with an appropriate subscription.

To use a different LLM provider (Azure OpenAI, Amazon Bedrock, OpenAI, or Google Vertex), configure a connector following the LLM connector documentation and update the inference_id parameter in the query to reference your configured connector.

Rule query

edit

from .alerts-security.*

// truncate timestamp to 5-minute window
| eval Esql.time_window_date_trunc = date_trunc(5 minutes, @timestamp)

// high severity alerts excluding system standard user.ids
| where kibana.alert.rule.name is not null and user.name is not null and kibana.alert.risk_score >= 73 and kibana.alert.workflow_status == "open" and
        not kibana.alert.rule.type in ("threat_match", "machine_learning") and
        not user.id in ("S-1-5-18", "S-1-5-19", "S-1-5-20", "0")

// group alerts by short time window and extract values of interest for alert triage
| stats Esql.event_module_distinct_count = COUNT_DISTINCT(event.module),
        Esql.user_name_distinct_count = COUNT_DISTINCT(user.name),
        Esql.rule_name_distinct_count = COUNT_DISTINCT(kibana.alert.rule.name),
        Esql.event_category_distinct_count = COUNT_DISTINCT(event.category),
        Esql.rule_risk_score_distinct_count = COUNT_DISTINCT(kibana.alert.risk_score),
        Esql.event_module_values = VALUES(event.module),
        Esql.rule_name_values = VALUES(kibana.alert.rule.name),
        Esql.message_values = VALUES(message),
        Esql.event_category_values = VALUES(event.category),
        Esql.event_action_values = VALUES(event.action),
        Esql.source_ip_values = VALUES(source.ip),
        Esql.destination_ip_values = VALUES(destination.ip),
        Esql.host_id_values = VALUES(host.id),
        Esql.agent_id_values = VALUES(agent.id),
        Esql.rule_severity_values = VALUES(kibana.alert.risk_score),
        Esql.user_name_values = VALUES(user.name) by Esql.time_window_date_trunc

// filter for alerts from different integrations with unique categories
| where Esql.event_module_distinct_count >= 2 and Esql.user_name_distinct_count >= 2 and Esql.event_category_distinct_count >= 2

// build context for LLM analysis
| eval users_list = MV_CONCAT(Esql.user_name_values, ",")

// LLM analysis
| eval instructions = "Analyze the provided user names and return a boolean value true if at least 2 of them are similar and they may belong to the same human identify or false if not, do not compare user names that may look like service accounts. If the list of users has more than 2 users and only 2 of them are similar consider this as true. Structure the output as follows: verdict=<verdict> confidence=<score> summary=<short reason max 500 words> without any other response statements on a single line."
| eval prompt = CONCAT("User identities extracted from different alerts: ", users_list, instructions)
| COMPLETION triage_result = prompt WITH { "inference_id": ".gp-llm-v2-completion"}

// parse LLM response
| DISSECT triage_result """verdict=%{Esql.verdict} confidence=%{Esql.confidence} summary=%{Esql.summary}"""

// filter for similar user values
| where TO_LOWER(Esql.verdict) == "true"
| keep Esql.*

« Control Panel Process with Unusual Arguments Creation of Hidden Files and Directories via CommandLine »