Skip to content

Nexus SecOps Reference Architecture

This document defines the reference architecture for an AI-Powered Security Operations Center. It describes the technology layers, their relationships, and design principles that underpin the Nexus SecOps benchmark.


Architecture Overview

flowchart TD
    subgraph SOURCES["Log Sources"]
        EP[Endpoints\nEDR / AV]
        NET[Network\nFirewall / NDR / DNS]
        ID[Identity\nIdP / AD / SSO]
        CLD[Cloud\nAWS / Azure / GCP]
        APP[Applications\nWeb / API / Email]
        OT[OT/ICS\nSensors via Diode]
    end

    subgraph COLLECT["Collection Layer"]
        AGT[Agents\n& Forwarders]
        API[API Collectors\n& Connectors]
        SYS[Syslog Receivers]
        BUF[Message Buffer\nKafka / Queue]
    end

    subgraph STORE["Storage and Processing"]
        SIEM[SIEM\nRule Engine + Correlation]
        DL[Data Lake\nLong-term + ML]
        HOT[Hot Storage\n0-90 days]
        WARM[Warm Storage\n90d-1yr]
        COLD[Cold Archive\n1yr+]
    end

    subgraph INTEL["Intelligence Layer"]
        TIP[Threat Intel\nPlatform]
        FEEDS[External Feeds\nISAC / OSINT]
        ENRICH[Enrichment\nGeo / ASN / CMDB]
    end

    subgraph DETECT["Detection Layer"]
        RULES[Rule Engine\nSigma / Native]
        ML[ML Models\nUEBA / Anomaly]
        CORR[Correlation\nMulti-event]
    end

    subgraph AUTO["Automation Layer"]
        SOAR[SOAR Platform]
        PB[Playbooks]
        GATE[Human Gates\nApproval Workflow]
    end

    subgraph PRESENT["Presentation Layer"]
        CASE[Case Management]
        DASH[SOC Dashboard]
        RPT[Reporting]
    end

    subgraph AI["AI Assistance Layer"]
        LLM[LLM Copilot]
        RAG[RAG Knowledge Base\nRunbooks / TI / History]
    end

    SOURCES --> COLLECT
    COLLECT --> BUF
    BUF --> STORE
    FEEDS --> TIP
    TIP --> INTEL
    ENRICH --> INTEL
    INTEL --> DETECT
    STORE --> DETECT
    DETECT --> AUTO
    AUTO --> PRESENT
    INTEL -.->|Contextual enrichment| PRESENT
    RAG --> LLM
    LLM -.->|Analyst assistance| PRESENT

Layer Descriptions

Collection Layer

Purpose: Reliably transport log data from sources to the storage and processing layer with minimal loss and latency.

Components:

Component Role Examples
Agents/Forwarders Installed on endpoints/servers; collect and forward logs Elastic Agent, Splunk UF, Filebeat
API Collectors Pull logs from cloud services and SaaS via API Custom scripts, Logstash, vendor connectors
Syslog Receivers Receive network device logs via UDP/TCP syslog rsyslog, syslog-ng
Message Buffer Decouple source from storage; absorb volume spikes Apache Kafka, Azure Event Hub, AWS Kinesis

Design principles: - Encrypt all transport (TLS 1.2+) — Nexus SecOps-003 - Buffer to prevent data loss during SIEM maintenance — Nexus SecOps-002 - Monitor delivery health; alert on gaps — Nexus SecOps-007 - Deploy agents at ≥98% coverage for managed endpoints — Nexus SecOps-006


Storage Layer

Purpose: Retain security logs in queryable form for detection and investigation, and in archived form for compliance.

Tiered Storage Model:

Tier Retention Access Speed Use Case
Hot (SIEM) 0–90 days Real-time Active detection, live investigation
Warm (Data Lake) 90 days–1 year Minutes Threat hunting, historical investigation
Cold (Archive) 1 year+ Hours Compliance, legal, forensics

SIEM vs. Data Lake:

SIEM Data Lake
Optimized for Real-time correlation, alerting Large-scale search, ML, long-term
Schema Rigid, pre-parsed Flexible, schema-on-read
Cost per GB High Low
Latency Sub-second Seconds to minutes
ML workloads Limited Native

Use a hybrid architecture: SIEM for real-time alerting (hot data), data lake for threat hunting, ML training, and retention (warm/cold data).


Intelligence Layer

Purpose: Enrich security events with contextual data that transforms raw logs into actionable intelligence.

Enrichment types:

Enrichment What It Adds Source
Asset context Host criticality, owner, OS, business unit CMDB / Asset inventory
User context Role, department, admin status, manager Active Directory / HR system
IP reputation Malicious/benign score, geo, ASN Threat intelligence feeds, VirusTotal
File reputation Known malware, file metadata Threat intelligence feeds
IOC matching Confirmed threat indicators SIEM IOC matching, TIP

Nexus SecOps controls: Nexus SecOps-018 (asset enrichment), Nexus SecOps-019 (IP enrichment), Nexus SecOps-026 (TI enrichment), Nexus SecOps-081–095 (CTI domain)


Detection Layer

Purpose: Identify threat behavior in log data through rules, correlation, and machine learning.

Detection engine types:

Type Mechanism Best For Nexus SecOps Control
Signature/rule Pattern matching on field values Known attack patterns Nexus SecOps-031, 032
Threshold Count-based triggers Brute force, scanning Nexus SecOps-031
Correlation Multi-event sequences across time Attack chains Nexus SecOps-046
Statistical anomaly Deviation from baseline Unknown threats, UEBA Nexus SecOps-044, 169
ML classification Trained model scores events Complex behavioral patterns Nexus SecOps-050

Detection content governance: All rules MUST follow the detection change control process (Nexus SecOps-033, Nexus SecOps-203). Rules are version-controlled in a git repository.


Automation Layer

Purpose: Accelerate response actions, reduce manual workload, and enforce consistent processes through SOAR playbooks.

Key design principles: - All high-impact actions (account disable, host isolation) MUST have human approval gates — Nexus SecOps-099 - Playbooks MUST be version-controlled and tested before deployment — Nexus SecOps-098, 100 - All automation actions MUST be logged with full audit trail — Nexus SecOps-102 - Rollback procedures MUST exist for all automated actions — Nexus SecOps-103

Automation tiers:

Tier Actions Human Gate?
Enrichment IP lookup, asset lookup, user context No
Notification Create ticket, send alert, assign to analyst No
Soft containment Revoke session, disable alert source No
Hard containment Disable account, isolate host, block IP Yes — required

Presentation Layer

Purpose: Surface actionable intelligence to analysts in the right format at the right time.

Components:

Component Audience Key Metrics
Alert queue dashboard Tier 1 analysts Queue depth, SLA status, open by severity
Investigation workbench Tier 2 analysts Timeline view, entity relationships, pivot tools
Case management Tier 2 / IR leads Incident timeline, evidence log, action tracker
Operational dashboard SOC Manager MTTD, MTTR, FP rate, SLA compliance
Executive dashboard CISO / Board Dwell time, critical incidents, risk posture

AI Assistance Layer

Purpose: Augment analyst productivity through LLM-powered copilots grounded in authoritative security knowledge.

Architecture:

flowchart LR
    Q[Analyst Query] --> EMB[Query Embedding]
    EMB --> VS[Vector Search]
    VS --> KB[(Knowledge Base\nRunbooks / TI Reports\nPlaybooks / History)]
    KB --> CTX[Retrieved Context]
    CTX --> PROMPT[Augmented Prompt]
    Q --> PROMPT
    PROMPT --> LLM[LLM API]
    LLM --> OUT[Response + Citations]
    OUT --> ANALYST[Analyst Review]

Key requirements: - All LLM deployments inventoried — Nexus SecOps-181 - Prompt injection defense implemented — Nexus SecOps-182 - RAG grounding for factual queries — Nexus SecOps-184 - PII filtered before external API submission — Nexus SecOps-186 - Human review required before acting on LLM output — Nexus SecOps-191


High Availability Design

Single points of failure to eliminate:

Component HA Approach
SIEM platform Active-active clustering or hot standby
Log collection buffer Multi-broker Kafka cluster; replication factor ≥ 3
SOAR platform Active-passive with automated failover
Log agents Agent-level buffer to disk; retry on connection failure
TIP/enrichment APIs Circuit breaker pattern; graceful degradation if API down

Nexus SecOps-007 requires alerting when log delivery stops. Monitor the monitoring infrastructure itself with an independent health check.


Security of the SOC Infrastructure

The SOC tools themselves are high-value targets. Apply these controls:

Risk Control
Unauthorized SIEM access MFA + RBAC; analysts see only authorized data — Nexus SecOps-114
SOAR credential theft Vault-based credential management; no hardcoded API keys — Nexus SecOps-104
Log tampering Write-once log storage; integrity hashing — Nexus SecOps-005
Lateral movement to SIEM Network segmentation; SOC tools in dedicated VLAN — Nexus SecOps-121
Supply chain attack on SIEM Vendor security assessment; software integrity verification
LLM data leakage PII filtering; on-premises LLM for sensitive data — Nexus SecOps-186, 196

Technology Selection Criteria

Layer Key Selection Criteria
SIEM Real-time query performance; rule language; native integrations; license model
Data Lake Cost per TB; query language (SQL/KQL/SPL); ML integration; retention flexibility
EDR Coverage (Windows/Linux/Mac/cloud); behavioral detection; isolation capability
SOAR Integration catalog breadth; human gate support; playbook versioning; audit log
TIP Feed quality; STIX/TAXII support; SIEM integration; indicator TTL management
LLM Copilot Grounding capability; PII controls; on-premises option; evaluation framework

See Data Pipeline Patterns | Integration Patterns | Zero Trust SOC Nexus SecOps controls: All 14 domains apply to this architecture.