Avispa OrbitSOC

Detections your SOC lead approves.

Agents propose rules from intel, coverage gaps and threat profiles, and a SOC lead puts them live.

Detection engineering
Reconnaissance
Resource Development
Initial Access
Execution
Persistence
Privilege Escalation
Defense Evasion
Credential Access
Discovery
Lateral Movement
Collection
Command and Control
Exfiltration
Impact

Proposed by the detection engineer agent

LSASS memory read by an unusual process

Why this rule
Intel
Two articles this month describe credential theft from LSASS.
Coverage
Credential Access is your thinnest covered tactic.
Telemetry
Process-access events: ready on Windows.
Confidence
Medium

A SOC lead deploys it.

Write once, in Sigma.

Content library

Sigma

title: LSASS memory read by an unusual process
logsource:
  product: windows
  category: process_access
detection:
  selection:
    TargetImage|endswith: '\lsass.exe'
    GrantedAccess|contains: '0x1010'
  filter:
    SourceImage|startswith: 'C:\Windows\System32\'
  condition: selection and not filter
level: high
tags: [attack.credential_access, attack.t1003.001]

Splunk SPL

sourcetype="XmlWinEventLog:Microsoft-Windows-Sysmon/Operational"
EventCode=10 TargetImage="*\\lsass.exe"
GrantedAccess="*0x1010*"
NOT SourceImage="C:\\Windows\\System32\\*"

Export forMicrosoft SentinelFalcon LogScaleGoogle SecOps

Library rules compile to Splunk SPL, and export for Microsoft Sentinel, Falcon LogScale and Google SecOps.

Every signal, remembered.

Learning

Lesson from an analyst

svc-reports signs in from RPT-SRV-02 at 01:00 UTC every night. That alone isn’t suspicious.

Reaches onlyTriage analystIdentity

Impact receipt once a SOC lead approves: Changed 3 investigations

InvestigationWhat it sawWhat changed
INV-2291svc-reports, 01:00 UTC, RPT-SRV-02Suspicious → benign
INV-2307svc-reports, 01:02 UTC, RPT-SRV-02Closed at triage
INV-2318svc-reports, 01:00 UTC, a new hostStill escalated

Analysts teach lessons and the AI proposes facts, and a SOC lead approves what it keeps. Approved knowledge reaches only the agents it names, and receipts show what it changed.

No detection deploys itself, no knowledge goes live unapproved, and learning never rewrites a closed case.

Bring your detection backlog to the demo.