Security

Where detection rules come from

Rules for new CVEs and open detection rules, converted to LPQL and tested before they reach you.

In short
LogPulse writes rules for new, exploited CVEs and converts open rules from SigmaHQ, Splunk security_content, Panther and Microsoft Sentinel to LPQL. Rules are tested on recordings of real attacks or on their own test events. Rules from the Detection Hub are published by a person and reach your stack within 15 minutes; all rules arrive switched off.

How a rule gets to you

Two ways a rule reaches your stack
Sigma rules ship with LogPulse itself. Everything else goes through the Detection Hub, where a person publishes it.
Where it starts
Sigmaopen rules
Splunk, Panther, Sentinelopen rules
New CVEsown rules
Build and test
Sigma bundleconverted, tested
Draft ruletests must pass
Deliver
Releasewith LogPulse
Detection Huba person publishes
Arrives switched off
Sigma rulecounted silently
Hub ruleyou switch it on
  1. Sigma to Sigma bundle
  2. Splunk, Panther, Sentinel to Draft rule
  3. New CVEs to Draft rule
  4. Sigma bundle to Release
  5. Draft rule to Detection Hub
  6. Release to Sigma rule
  7. Detection Hub to Hub rule

Each open repository is pinned to a commit. A check compares the pin with upstream, so new upstream rules go through the same steps. Converted rules keep the original author, the licence and a note that LogPulse converted them.

From a CVE to a rule

Next to open rules, LogPulse writes its own rules for new vulnerabilities. New CVEs are triaged within one hour during business hours, so a rule for an exploited CVE can be ready the same day. It reaches you once a person has published it.

From a new CVE to your stack
Feeds
CVE feedsNVD, KEV, EPSS, news
Triage
Triageneeds a rule?
Exploit pathadvisories, PoCs
Build, test, publish
Rulesees the attack
Version checkfinds vulnerable hosts
Mitigationpatch or workaround
Within 15 minutes
Your stackrules off, checks on
  1. CVE feeds to Triage
  2. Triage to Exploit path
  3. Exploit path to Rule
  4. Exploit path to Version check
  5. Exploit path to Mitigation
  6. Rule to Your stack
  7. Version check to Your stack
  8. Mitigation to Your stack

A CVE asks for a rule when it is in the CISA list of exploited vulnerabilities (KEV), has an EPSS score of 0.1 or higher, is in the security news, runs at LogPulse customers, or affects widely used, often internet-facing software such as Apache, nginx, Exchange, NetScaler, FortiOS or Ivanti Connect Secure. Then we find out how the attack shows up in logs, mostly the request path, from the vendor advisory and public exploits. A CVE then gets one of these states:

StateMeaning
CoveredA published rule already sees the CVE being exploited; the CVE is linked to it.
In progressA rule is being written and tested.
Rule neededA rule can be written now and is on the list.
BlockedDetectable in principle, but not yet: the exploit is in the request body, there is no log source for the product, no public request details exist, or it uses a protocol LogPulse does not log. The reason is kept.
Not detectableLogs cannot show it, such as cross-site scripting or another client-side bug. The reason is kept.

One rule covers at most 50 CVEs. Where a log cannot show the attack, a version check by the agent can still find the hosts that run the vulnerable version, and a mitigation says what to patch. All three are published by a person. On your stack a rule arrives switched off; a version check arrives switched on and can be switched off.

Note: Which CVEs LogPulse detects is public: logpulse.io/cve-coverage lists the processed CVEs with their status (rule, agent check, in development or not detectable), and /cve/<id> has a page per CVE. Only published content counts; how a rule detects stays private.

Open sources

RepositoryLicenceReaches youWhat we use and how it is tested
SigmaHQDetection Rule License 1.1With a LogPulse releaseWindows rules run on SigmaHQ’s own attack samples and on the attack_data recordings
Splunk security_contentApache-2.0Detection HubEndpoint rules (processes, registry, files); a rule has to fire on its own attack recording
PantherApache-2.0Detection HubAWS CloudTrail rules; a rule has to fire exactly where Panther’s own test logs say it should
Microsoft SentinelMITDetection HubKQL rules for CloudTrail and Defender data; the converter builds test events from the KQL, because the repository has none

The converters only translate what they can translate exactly. A rule they cannot translate is refused with a reason; it is not guessed. Converted rules use an index per kind of data, such as windows or audit, and move to the CIM field names.

Tests

TestWhat it proves
attack_data gateSysmon and PowerShell recordings of real attacks go through a copy of the Windows parser and every Windows rule we ship. A technique that was detected and no longer is fails the build.
Own recordingA rule converted from Splunk runs on the recording that belongs to it. Only rules that fire on it go to the hub.
Vendor test logsA rule converted from Panther has to fire exactly where Panther’s own test logs say it should, and stay quiet on the others.
Rule tests in the hubAn event that must fire and a normal event that must not. A rule cannot be published while a test fails.
Note: The gate also catches rules that fire on the wrong data. It found a macOS rule and an antivirus rule that matched Windows events; both are now scoped to their own source.

On your stack

Rules from the Detection Hub. Your stack fetches the published catalog every 15 minutes and checks it against a signed manifest before it replaces anything. Then:

  • a new rule is added switched off;
  • a rule you did not edit gets the new version and keeps your parameter values;
  • a rule you edited stays as it is and shows that an update is available;
  • a rule the hub retires keeps running where it is on, with a warning, and disappears where it is off.

Sigma rules. They arrive with a LogPulse release, also switched off. Every 15 minutes LogPulse counts silently how often a switched-off Sigma rule would have matched your data; nothing is raised. A rule that passed its attack test, has high or critical severity and stayed quiet for a week switches itself on in monitor mode for a while: it records matches without raising notables. After that it works like any rule. A rule you switched off by hand stays off, and you can turn this behaviour off for your organisation.

In the app, open Detection Hub, Library and filter on New to see the rules that just arrived. See Security Monitoring for how detections become notables.

Frequently asked questions

Where do the LogPulse detection rules come from?
From LogPulse itself, including rules for new CVEs, and from open detection repositories: SigmaHQ, Splunk security_content, Panther and Microsoft Sentinel. Each repository is pinned to a commit, with its licence and attribution kept.
How are converted rules tested?
Recordings of real attacks from splunk/attack_data go through a copy of the Windows parser and then through every Windows rule we ship. A technique that was detected and stops being detected fails the build. Rules from Panther run on Panther’s own test logs, and every rule in the hub needs passing tests before it can be published.
Are new detection rules switched on automatically?
Rules from the Detection Hub are not: a person publishes them, your stack fetches them within 15 minutes and adds them switched off. Sigma rules also arrive switched off and are counted silently first. A tested Sigma rule of high or critical severity that stays quiet for a week switches itself on in monitor mode, unless you turn that setting off.
How fast does LogPulse add a rule for a new CVE?
New CVEs are triaged within one hour during business hours. CVEs in the CISA KEV list, with an EPSS score of 0.1 or higher, in the news or in widely used software go first. Once a person has published the rule, your stack has it within 15 minutes, switched off.
Which CVEs does LogPulse detect?
See logpulse.io/cve-coverage: every processed CVE with its status (rule, agent version check, in development or not detectable) and a page per CVE.
What happens when a rule I edited gets an update?
Your version stays as it is and the rule shows that an update is available. Rules you did not edit get the new version and keep your parameter values.
What if a rule cannot be converted?
The converter refuses it and gives a reason, such as a Splunk command that has no exact LPQL equivalent. It does not guess.