Skip to content

Hermes Autonomous Cyber Threat Intelligence Engine

Density:
โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”
โ”‚ HERMES CTI ENGINE DATA FLOW PIPELINE โ”‚
โ”‚ โ”‚
โ”‚ RAW OPEN-SOURCE SOURCES (5 Pillars: KEV, NVD, Shadowserver, Talos, etc.) โ”‚
โ”‚ โ”‚ โ”‚
โ”‚ โ–ผ โ”‚
โ”‚ [BaseCollector: fetch() -> SHA-256 Checksum -> Immutable Raw Cache] โ”‚
โ”‚ โ”‚ โ”‚
โ”‚ โ–ผ โ”‚
โ”‚ [LicensingGuard: redistribution_allowed, commercial_use, derived_data] โ”‚
โ”‚ โ”‚ โ”‚
โ”‚ โ–ผ โ”‚
โ”‚ [AliasResolver: Collapsing Actor & Malware Aliases to Canonical IDs] โ”‚
โ”‚ โ”‚ โ”‚
โ”‚ โ–ผ โ”‚
โ”‚ [QualityChecker: first_seen <= last_seen, IOC Syntax, Evidence Provenance] โ”‚
โ”‚ โ”‚ โ”‚
โ”‚ โ–ผ โ”‚
โ”‚ [Epistemic Separation: FACT -> OBSERVATION -> INFERENCE -> PREDICTION] โ”‚
โ”‚ โ”‚ โ”‚
โ”‚ โ–ผ โ”‚
โ”‚ [Static Compiler: REST API (/api/cti/*), STIX 2.1, Daily Snapshots] โ”‚
โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜

The Hermes CTI Engine implements a Static-First Threat Intelligence Architecture. It operates on the principle that threat analysis must be reproducible, transparent, and completely free of opaque server database dependencies:

  • Zero Runtime Database: Intelligence artifacts are persisted as version-controlled, immutable JSON and STIX 2.1 entities compiled at build time.
  • Continuous Ingestion: Collectors systematically query and normalize data from authoritative cybersecurity feeds.
  • Cryptographic Traceability: Every raw payload is preserved under data/raw/<source>/YYYY/MM/DD/ alongside its cryptographic SHA-256 manifest and acquisition timestamp.

Hermes structures open-source intelligence collection across five distinct, authoritative pillars:

Pillar 1: Government & Standards

Authoritative baseline repositories and standards bodies:

  • CISA KEV (Known Exploited Vulnerabilities catalog)
  • CERT-FR (French National Cybersecurity Agency alerts)
  • NIST NVD (CVE scoring and CWE enumeration)
  • FIRST EPSS (Empirical Exploit Prediction Scoring System)
  • GitHub Security Advisories (Open-source package ecosystems)

Pillar 2: Technical Telemetry & Scanners

Global Internet-wide scanning and honeypot sensor telemetry:

  • Shadowserver Foundation (Exposed honeypot telemetry and active scanning)
  • CrowdSec Network (Collaborative distributed IP blocklists)
  • VulnCheck KEV & Exploits (Pre-NVD exploit weaponization tracking)
  • Patchstack (CMS and Web application exploit monitoring)

Pillar 3: Open Threat Feeds

High-fidelity threat feeds and malware telemetry:

  • abuse.ch URLhaus (Active malicious URL and delivery tracking)
  • abuse.ch ThreatFox (Crowdsourced indicator of compromise sharing)

Pillar 4: Knowledge Bases & Ontologies

Standardized attack behavior and taxonomy ontologies:

  • MITRE ATT&CK (Enterprise adversary tactics, techniques, and procedures)
  • MITRE ATLAS (Adversarial Threat Landscape for Artificial-Intelligence Systems)

Pillar 5: Security Vendor Threat Intelligence

Deep investigative reporting and telemetry from leading research labs:

  • Mandiant / Google Cloud Threat Intelligence (Nation-state campaigns, UNC threat clusters)
  • Cisco Talos Intelligence (Implant reverse-engineering, C2 infrastructure)
  • Microsoft Threat Intelligence (MSTIC) (Cloud identity abuse, adversary tracking)
  • Google Project Zero (In-the-wild zero-day exploit analysis)
  • Trend Micro ZDI & Kaspersky GReAT (Vulnerability disclosures and APT telemetry)
  • ANSSI & Vendor Advisories (Critical security bulletins from Oracle, Cisco, etc.)

A foundational tenet of Hermes is that AI is never an authoritative source of facts. The CTI engine enforces an unambiguous epistemic ladder:

  1. FACT: Information directly corroborated by an authoritative primary source (e.g., CISA KEV publication, cryptographically signed advisory).
  2. OBSERVATION: A deterministic, mechanically computed finding or telemetry delta (e.g., weaponization wave detected, honeypot spike of 340%).
  3. INFERENCE: Analytical hypotheses and reasoning derived from cross-source correlation (e.g., infrastructure reuse between UNC6240 and ransomware affiliates).
  4. PREDICTION: Forward-looking, falsifiable forecasts with defined resolution oracles and Brier score calibration.
  5. DECISION: Actionable gating directives (PATCH, MITIGATE, ISOLATE, MONITOR) tied directly to observed threat velocity.

Threat actors and malware families are notoriously fragmented across commercial vendors who assign proprietary nomenclature. The Hermes AliasResolver engine maps diverse vendor labels to canonical Hermes entities:

Canonical IDCanonical NameResolved Vendor Aliases
ACTOR-UNC6240ShinyHuntersUNC6240, ShinyHunters, Shiny Corps
ACTOR-APT28APT28Fancy Bear, Sofacy, Sednit, Strontium, Forest Blizzard
ACTOR-APT29APT29Cozy Bear, Nobelium, Midnight Blizzard, The Dukes
ACTOR-VOID-ARACHNEVoid ArachneStorm-1944, CyberArachne Group
ACTOR-SCATTERED-SPIDERScattered SpiderUNC3944, 0ktapus, Starfrost, Octo Tempest
MAL-CLOSEDQUORUMCLOSEDQUORUMClosed Quorum, cairn-c2-implant
MAL-MESHCENTRALMeshCentralmeshagent, Mesh Central

Hermes respects intellectual property and vendor redistribution constraints via the LicensingGuard:

  • Public vs Derived: For sources with restrictive redistribution licenses (e.g., commercial research reports from Mandiant, Cisco Talos, Kaspersky), Hermes never republishes verbatim raw text.
  • Derived Epistemic Graph: Extracted claims, graph relationships, confidence metrics, and MITRE TTP alignments are permissible derived intelligence and are exported with mandatory attribution.
  • Public Domain / CC Sources: Sources permitting raw redistribution (CISA KEV, NVD, abuse.ch) are fully queryable and redistributable.

To prevent misinformation and hallucination, evidence is evaluated across a strict six-tier hierarchy:

  1. Tier 1 โ€” Primary Source: Authoritative standards bodies and sovereign CERTs (SRC-CISA-KEV, SRC-CERT-FR, SRC-NVD).
  2. Tier 2 โ€” Official Vendor: Primary product vendor security advisories (SRC-ORACLE-SECURITY, SRC-CISCO-SECURITY).
  3. Tier 3 โ€” Original Research: First-hand reverse engineering and forensic investigations (SRC-MANDIANT-GOOGLE, SRC-CISCO-TALOS, SRC-GOOGLE-PROJECT-ZERO).
  4. Tier 4 โ€” Independent Corroboration: Telemetry networks and scanning foundations (SRC-SHADOWSERVER, SRC-CROWDSEC, SRC-VULNCHECK).
  5. Tier 5 โ€” Secondary Reporting & Feeds: Aggregated indicator databases and community lists (SRC-ABUSE-CH-*, SRC-FIRST-EPSS).
  6. Tier 6 โ€” Discovery-Only: Search engine results and unverified social signals (never accepted as standalone proof).
  • Atomic Claims: Complex threat narratives are atomized into structured Claim objects tagged with epistemic states (FACT, OBSERVATION, INFERENCE, PREDICTION, DECISION).
  • Epistemic Divergence: When authoritative sources disagree (e.g., syndicate organizational cohesion vs. access broker leasing), Hermes never forces an artificial consensus. Both claims are preserved with their respective evidentiary chains and flagged as DIVERGENT.

Raw observations are synthesized into structured analytic models:

  • Actor Analysis: Evaluates activity level, campaign frequency, TTP diversity, malware diversity, sector expansion, and vulnerability exploitation counts.
  • Campaign Analysis: Tracks campaign duration, targeted infrastructure, victim demographics, and weaponization pace.
  • Malware Analysis: Classifies dual-use tools (MAL-MESHCENTRAL), traditional backdoors (MAL-COBALT-STRIKE), and autonomous AI implants (MAL-CLOSEDQUORUM).
  • Infrastructure Analysis: Aggregates IP reputation, Tor exit nodes, active C2 domains, and payload SHA-256 signatures with full licensing compliance.
  • TTP Analysis & Evolution: Tracks MITRE ATT&CK and MITRE ATLAS technique velocity, identifying techniques with accelerating adoption (+34% 30-day delta).

Threat actors and campaigns are not static risk scores; they follow continuous temporal trajectories:

Current State โ”€โ”€โ”€โ–บ Velocity โ”€โ”€โ”€โ–บ Acceleration โ”€โ”€โ”€โ–บ Archetype โ”€โ”€โ”€โ–บ Forecast

Hermes classifies trajectories into 11 Explainable Canonical Archetypes:

ArchetypeOperational CriteriaStrategic Implication
EMERGINGFirst seen $\le$ 45 days, surging initial velocityRapid onset monitoring required
ACCELERATINGAcceleration index $> 15$, multiple active campaignsImmediate perimeter hardening
STABLE_HIGHLong-term active syndicate with sustained baselineContinuous defensive posture
DECLININGNegative velocity, infrastructure takedownsDeprioritized operational alert
REACTIVATINGDormant actor exhibiting sudden new IOCsResumed campaign warning
EXPANDINGTargeting $\ge 4$ new sectors or geographic regionsCross-industry perimeter alerts
CONTRACTINGNarrowing targeting footprint to specific verticalsFocused vertical defense
TARGET_SHIFTPivot from legacy software to AI/Cloud perimetersArchitectural review needed
TTP_SHIFTAdoption of novel evasion or execution techniquesDetection rule engineering
INFRASTRUCTURE_EXPANSIONHigh volume of newly registered C2 domains/IPsActive blocklist updating
CAMPAIGN_BURSTShort-horizon mass exploitation waveEmergency patch gating

The CTI Engine continuously audits trajectories to flag Inflection Events (INFLECTION_EVENT):

  • Activity Spike: Rapid velocity acceleration across telemetry sensors.
  • Multi-Vulnerability Exploitation: Adversaries chaining multiple zero-day or KEV vulnerabilities (e.g., CVE-2026-35273 + CVE-2026-96560).
  • Novel Paradigm Shift: Discovery of operator-less autonomous AI C2 consensus channels (AML.T0043).

The CTI pipeline compiles static, cached JSON REST endpoints under /api/cti/:

REST EndpointPurposeSchema / Specification
/api/cti/sources.json22 Curated Authoritative Sources Registrysources.schema.json
/api/cti/actors.jsonCanonical Threat Actor Profiles & Aliasesactor.schema.json
/api/cti/campaigns.jsonActive Threat Campaigns & Targetingcampaign.schema.json
/api/cti/malware.jsonMalware Families, Dual-Use Tools & AI Implantsmalware.schema.json
/api/cti/indicators.jsonNormalized IOCs & Lifecycle Statusindicator.schema.json
/api/cti/claims.jsonEpistemic Claims & Evidence Pointersclaim.schema.json
/api/cti/contradictions.jsonDivergent Vendor Assessments & Tradecraftcontradiction.schema.json
/api/cti/observations.jsonTemporal Event Observationsobservation.schema.json
/api/cti/relationships.jsonKnowledge Graph Relational EdgesGraph Specification
/api/cti/trajectories.jsonThreat Trajectories & 11 ArchetypesSections 28-30
/api/cti/inflections.jsonAutomated Inflection Event DetectionSection 31
/api/cti/ttps.jsonTTP Frequency & Evolution AnalyticsSection 27
/api/cti/threat_landscape.jsonLayer 6 Global Landscape ProductSection 32
/api/cti/actors_analysis.jsonMulti-Dimensional Actor IntelligenceSection 23
/api/cti/quality_report.jsonAutomated Invariant & Syntax AuditInvariant Suite
/api/cti/snapshots/latest.jsonLatest Daily CTI Snapshot BundleTemporal Archive