🛡️ SHACL Profiles per Trust Level

Below is a tiered SHACL profile system that lets you expose the same graph to different consumers without changing the data, only the validation contract.

Think of it as semantic zero-trust.

(SOC / Exec / AI)

Design Principle

Trust level ≠ data visibility Trust level = which semantics are allowed to survive validation

Each profile:

⛔ forbids certain edges/classes
✅ enforces different minimum guarantees
🔎 controls what kind of reasoning is possible

0. Trust Levels Overview

Profile	Audience	Goal	Risk Tolerance
SOC	Analysts, IR, DFIR	Tactical accuracy	🔴 Zero
EXEC	CISO, Board, GRC	Decision clarity	🟠 Low
AI	RAG / Agents / LLMs	Hallucination safety	🔴 Zero

⚠️ AI and SOC are stricter than EXEC, but in different ways.

🔴 TRUST PROFILE: SOC

“Only what we can prove”

SOC Profile Philosophy

Observational truth only
No speculation
No aggregation shortcuts
Evidence is mandatory

SH-SOC-01 — Incidents Must Be Evidence-Backed

inc:SOCIncidentEvidenceShape
  a sh:NodeShape ;
  sh:targetClass inc:Incident ;
  sh:property [
    sh:path inc:observed ;
    sh:minCount 1 ;
  ] ;
  sh:property [
    sh:path inc:supported_by ;
    sh:minCount 1 ;
    sh:message "SOC incidents must be supported by evidence." ;
  ] .

SH-SOC-02 — Observed Techniques Require High Confidence

inc:SOCObservedTechniqueConfidenceShape
  a sh:NodeShape ;
  sh:targetClass inc:ObservedTechnique ;
  sh:property [
    sh:path inc:confidence ;
    sh:in ("HIGH") ;
    sh:message "SOC view requires HIGH confidence observations only." ;
  ] .

SH-SOC-03 — No Risk or Attribution Allowed

soc:NoRiskOrAttributionShape
  a sh:NodeShape ;
  sh:targetClass inc:Incident ;
  sh:not [
    sh:or (
      [ sh:path risk:evaluates ; sh:minCount 1 ]
      [ sh:path ta:attributes ; sh:minCount 1 ]
    )
  ] ;
  sh:message "SOC view must not include Risk or Attribution." .

✅ SOC Guarantees

Court-defensible
DFIR-safe
Zero speculation

🟠 TRUST PROFILE: EXEC

“Enough truth to decide”

EXEC Profile Philosophy

Aggregation allowed
Subjectivity allowed (explicit)
Attribution allowed (summarized)
Evidence optional

SH-EXEC-01 — Risk Required for Exposure

risk:ExecRiskRequiredShape
  a sh:NodeShape ;
  sh:targetClass sec:Vulnerability ;
  sh:property [
    sh:path risk:considered_in ;
    sh:minCount 1 ;
    sh:message "Exec view requires Vulnerabilities to be contextualized by Risk." ;
  ] .

SH-EXEC-02 — Attribution Must Be Abstracted

exec:AbstractAttributionOnlyShape
  a sh:NodeShape ;
  sh:targetClass ta:ThreatActor ;
  sh:property [
    sh:path ta:actor_type ;
    sh:in ("GROUP" "STATE" "UNKNOWN") ;
    sh:message "Exec attribution must be abstracted (no individual actors)." ;
  ] .

SH-EXEC-03 — Evidence Optional, Rationale Mandatory

risk:ExecRationaleShape
  a sh:NodeShape ;
  sh:targetClass risk:RiskAssessment ;
  sh:property [
    sh:path risk:rationale ;
    sh:minCount 1 ;
    sh:message "Exec decisions require rationale." ;
  ] .

✅ EXEC Guarantees

Decision-focused
Board-safe
No technical overload

🔴 TRUST PROFILE: AI (RAG / Agents)

“Nothing that can hallucinate”

AI Profile Philosophy

Deterministic traversal only
No subjective nodes without anchors
No unbounded fan-out
No implicit joins

SH-AI-01 — Only Approved Traversal Anchors

ai:ApprovedAnchorsShape
  a sh:NodeShape ;
  sh:targetClass sec:Technique ;
  sh:property [
    sh:path sec:belongs_to ;
    sh:minCount 1 ;
  ] ;
  sh:property [
    sh:path sec:mitigated_by ;
    sh:minCount 1 ;
    sh:message "AI queries require Technique anchors with mitigation context." ;
  ] .

SH-AI-02 — No Direct Cross-Extension Jumps

ai:NoCrossExtensionJumpShape
  a sh:NodeShape ;
  sh:targetClass inc:Incident ;
  sh:not [
    sh:property [
      sh:path risk:evaluates ;
      sh:minCount 1 ;
    ]
  ] ;
  sh:message "AI must not traverse Incident → Risk directly." .

SH-AI-03 — Attribution Requires Confidence Threshold

ai:AttributionConfidenceThresholdShape
  a sh:NodeShape ;
  sh:targetClass ta:AttributionClaim ;
  sh:property [
    sh:path ta:confidence_score ;
    sh:minInclusive 0.7 ;
    sh:message "AI attribution requires confidence ≥ 0.7." ;
  ] .

SH-AI-04 — No Unanchored Risk

ai:AnchoredRiskOnlyShape
  a sh:NodeShape ;
  sh:targetClass risk:RiskScenario ;
  sh:property [
    sh:path risk:considers ;
    sh:class sec:Technique ;
    sh:minCount 1 ;
    sh:message "AI risk reasoning must anchor to Core Technique." ;
  ] .

✅ AI Guarantees

Hallucination-resistant
Deterministic answers
Auditable reasoning paths

🧩 How to Deploy This (Recommended)

1. One Graph, Multiple SHACL Sets

/shacl/
 ├── core.ttl
 ├── incident.ttl
 ├── risk.ttl
 ├── threatactor.ttl
 ├── trust-soc.ttl
 ├── trust-exec.ttl
 └── trust-ai.ttl

2. Validate by Context

SOC UI → core + incident + trust-soc
Exec Dashboard → core + risk + trust-exec
RAG Pipeline → core + all + trust-ai