Source overview

elastic/detection-rules

Official Elastic SIEM detection rules — TOML wrapping EQL/KQL/Lucene queries, mapped to ATT&CK.

elastic-detection-rules · Elastic TOML · upstream repo ↗ · Elastic License 2.0

Rules indexed 2237 8 atomic-IOC hidden · include 130 deprecated hidden · include
ATT&CK techniques 449
Newest rule 2026-11-19 checked 2026-11-22

How detections work here

Elastic's own detection content, run as scheduled queries against Elasticsearch by the Detection Engine. This is the model most of this index assumes: a query, a lookback window, an interval.

Do not confuse it with elastic-protections-artifacts, which ships inside the Elastic Defend agent and runs on the endpoint.

How rules are written

TOML, with [metadata] and [rule] tables. The important field is language, because this source carries several query languages — kuery, eql, esql — and the rule's type selects which engine runs it. That is why this is one of the few sources here with a meaningful content-type breakdown: query, eql, esql, threshold, new_terms, machine_learning and threat_match are genuinely different detection mechanisms, not labels.

index declares the indices searched and from the lookback (now-9m). risk_score is a 1–100 integer, which we preserve and also bucket into this site's five-level severity — both are shown on a rule's page.

Reading a rule on this site

Two prose fields map to distinct places, and the distinction is worth knowing: setup becomes the implementation guide (what you must configure before the rule works), while note becomes triage notes (what to do when it fires). Elastic's note fields are unusually thorough investigation guides, often longer than the rule itself.

ATT&CK comes from the [[rule.threat]] blocks and is generally complete.

Gotchas

  • type = "machine_learning" rules have no readable query — they reference a trained job by name. The rule tells you what it alerts on, not how it decides.
  • Every rule is scoped to specific index patterns and Elastic integrations. A rule is only deployable if you are shipping that integration's data in that shape.
  • Some investigation guides are noted upstream as generated with AI assistance and reviewed. Read them as a starting point.

Rungs

Reaches TTPs — 201 rules, 9.0% of the ruleset. A rung is claimed at 20 rules and 1.0% of the source, so that one rule cannot speak for a small ruleset and a rounding error cannot speak for a large one.

Shares are of all 2245 live rules in this source, including the atomic-IOC and non-detection content the count above hides by default. The pending band is drawn to scale for the same reason: it is the width of what no classifier has read, and a bar without it would look more complete than the index is. What the rungs mean.


Severity

Status


Platforms

endpoint 998 windows 457 sentinel_one_cloud_funnel 438 crowdstrike 328 m365_defender 262 aws 261 system 253 azure 164 +56 more

Content types

eql 979 query 487 esql 382 new_terms 251 machine_learning 109 threshold 29


Recently modified

all →
M365 Identity Login from Atypical Region 2026-11-19
Endpoint Discovery Commands Followed by IMDS Requests 2026-11-18
Entra ID Potential Passkey Downgrade 2026-11-18
Java HotSpot Fatal Error Log Created 2026-11-18
Kubernetes Pod Creation Using Common Debug or Base Images 2026-11-18

Detection rules belong to the projects that publish them and remain under their own licenses. This site indexes and links to them; it claims no rights in them.