Skip to content

KNOWLEDGE-GRAPH.md

Latest commit

 

History

History
578 lines (459 loc) · 14.9 KB

KNOWLEDGE-GRAPH.md

File metadata and controls

578 lines (459 loc) · 14.9 KB
Raw

Knowledge Graph: Australian AI Security Frameworks

Visual mapping of framework relationships

This document provides visual representations of how Australian AI security frameworks connect and relate to each other.

Framework Hierarchy

graph TB
    subgraph Federal["Federal Level"]
        ISM[Information Security Manual]
        PSPF[Protective Security Policy Framework]
        SOCI[SOCI Act / CIRMP]
        DTA[DTA AI Policy v2.0]
        
        subgraph ACSC["ACSC Guidance"]
            ACSC1[Engaging with AI]
            ACSC2[Deploying AI Securely]
            ACSC3[Secure AI Development]
            ACSC4[AI Data Security]
            ACSC5[AI Supply Chain]
            ACSC6[Content Credentials]
            ACSC7[Frontier Models]
        end
        
        subgraph DISR["DISR/NAIC"]
            AI6[AI6 Guidance]
            ETHICS[AI Ethics Principles]
            NAIP[National AI Plan]
            AISI[AI Safety Institute]
        end
    end
    
    subgraph Sector["Sector Regulators"]
        APRA[APRA CPS 234/230]
        ASIC[ASIC Report 798]
        ESAFE[eSafety Standards]
        OAIC[OAIC Privacy]
        TGA[TGA SaMD]
    end
    
    subgraph National["Cross-Jurisdictional"]
        NFAI[National Framework for AI Assurance]
    end
    
    subgraph States["State/Territory"]
        NSW[NSW AIAF]
        VIC[VIC GenAI Guideline]
        QLD[QLD FAIRA]
        WA[WA AI Policy]
        SA[SA AI Ethics]
        ACT[ACT AI Policy]
        NT[NT Assurance Framework]
        TAS[TAS Guidance]
    end
    
    %% Federal relationships
    ISM --> DTA
    PSPF --> DTA
    ISM --> ACSC1
    ISM --> ACSC2
    
    %% National framework relationships
    NFAI --> NSW
    NFAI --> VIC
    NFAI --> QLD
    NFAI --> WA
    NFAI --> SA
    NFAI --> ACT
    NFAI --> NT
    NFAI --> TAS
    
    %% Ethics principles flow
    ETHICS --> NFAI
    ETHICS --> DTA
    ETHICS --> NSW
    ETHICS --> VIC
    ETHICS --> QLD

    %% National AI Plan relationships
    NAIP --> AISI
    AISI -.-> ACSC1
    NAIP --> ETHICS
    NAIP --> DTA

    %% ISM referenced by states
    ISM -.-> NSW
    ISM -.-> VIC
    ISM -.-> QLD
    ISM -.-> WA
    
    %% Sector flows
    ISM -.-> APRA
    ISM -.-> ASIC
    
    classDef mandatory fill:#ff6b6b,stroke:#333,stroke-width:2px
    classDef voluntary fill:#4ecdc4,stroke:#333,stroke-width:2px
    classDef sector fill:#ffe66d,stroke:#333,stroke-width:2px
    classDef state fill:#95e1d3,stroke:#333,stroke-width:2px
    
    class ISM,PSPF,SOCI,DTA mandatory
    class ACSC1,ACSC2,ACSC3,ACSC4,ACSC5,ACSC6,ACSC7,AI6,ETHICS,NAIP,AISI voluntary
    class APRA,ASIC,ESAFE,OAIC,TGA sector
    class NSW,VIC,QLD,WA,SA,ACT,NT,TAS state
Loading

Document Relationships

Federal Framework Dependencies

graph LR
    subgraph Core["Core Frameworks"]
        ISM[ISM]
        PSPF[PSPF]
    end
    
    subgraph AI_Controls["AI-Specific"]
        ISM1923[ISM-1923<br/>OWASP LLM]
        ISM1924[ISM-1924<br/>Adversarial Detection]
        ISM2072[ISM-2072<br/>Model Formats]
        PSPF001[PSPF 001-2025<br/>GenAI Rules]
        DEEPSEEK[PSPF Direction<br/>DeepSeek Ban]
    end
    
    subgraph External["External References"]
        OWASP[OWASP Top 10 LLM]
        E8[Essential Eight]
    end
    
    ISM --> ISM1923
    ISM --> ISM1924
    ISM --> ISM2072
    PSPF --> PSPF001
    PSPF --> DEEPSEEK
    
    ISM1923 --> OWASP
    ISM --> E8
    
    classDef control fill:#ff9999,stroke:#333
    classDef external fill:#99ccff,stroke:#333
    
    class ISM1923,ISM1924,ISM2072,PSPF001,DEEPSEEK control
    class OWASP,E8 external
Loading

State Framework Maturity

graph TB
    subgraph Comprehensive["Comprehensive"]
        NSW[NSW<br/>AIAF + Ethics + Review Committee]
        WA[WA<br/>Policy v2 + Framework + Advisory Board]
    end
    
    subgraph Developing["Developing"]
        VIC[VIC<br/>GenAI Guideline + VPDSF]
        QLD[QLD<br/>FAIRA + IS18]
        ACT[ACT<br/>Policy + Framework + AIAG]
        SA[SA<br/>Ethics Policy + Office for AI]
    end

    subgraph Basic["Basic"]
        NT[NT<br/>Assurance Framework]
    end
    
    subgraph Minimal["Minimal"]
        TAS[TAS<br/>Guidance only]
    end
    
    %% Cross-references
    NSW -.->|referenced by| TAS
    NSW -.->|model for| ACT
    
    classDef comp fill:#2ecc71,stroke:#333,stroke-width:2px
    classDef dev fill:#f39c12,stroke:#333,stroke-width:2px
    classDef basic fill:#e74c3c,stroke:#333,stroke-width:2px
    classDef minimal fill:#95a5a6,stroke:#333,stroke-width:2px
    
    class NSW,WA comp
    class VIC,QLD,ACT,SA dev
    class NT basic
    class TAS minimal
Loading

Compliance Pathways

Commonwealth Entity

flowchart TD
    START[Commonwealth Entity<br/>using AI] --> ISM{ISM Controls<br/>Required?}
    ISM -->|Yes| ISM_COMPLY[Implement ISM-1923,<br/>ISM-1924, ISM-2072]
    ISM_COMPLY --> PSPF{Processing<br/>OFFICIAL data?}
    PSPF -->|Yes| PSPF_CHECK{Provider<br/>approved?}
    PSPF_CHECK -->|OpenAI/Anthropic| PROCEED[Proceed with<br/>logging]
    PSPF_CHECK -->|Other| FOCI[FOCI Assessment<br/>Required]
    PSPF_CHECK -->|DeepSeek| BANNED[Prohibited]
    PSPF -->|No| DTA[DTA Policy<br/>Compliance]
    
    PROCEED --> DTA
    FOCI -->|Pass| PROCEED
    FOCI -->|Fail| BANNED
    
    DTA --> CAIO[Designate Chief<br/>AI Officer]
    CAIO --> RISK{Risk Level?}
    RISK -->|High| OVERSIGHT[Submit to<br/>Oversight Committee]
    RISK -->|Low/Medium| DOCUMENT[Document and<br/>Monitor]
    
    OVERSIGHT --> TRANSPARENCY[Publish Transparency<br/>Statement]
    DOCUMENT --> TRANSPARENCY
    
    classDef start fill:#3498db,stroke:#333
    classDef decision fill:#f1c40f,stroke:#333
    classDef action fill:#2ecc71,stroke:#333
    classDef stop fill:#e74c3c,stroke:#333
    
    class START start
    class ISM,PSPF,PSPF_CHECK,RISK decision
    class ISM_COMPLY,PROCEED,DTA,CAIO,OVERSIGHT,DOCUMENT,TRANSPARENCY,FOCI action
    class BANNED stop
Loading

Critical Infrastructure Owner

flowchart TD
    START[CI Asset Owner<br/>deploying AI] --> SOCI{SOCI Act<br/>Applies?}
    SOCI -->|Yes| CIRMP[Include in CIRMP<br/>Cyber Hazard Vector]
    
    CIRMP --> RISK_CAT[Categorise AI Risks]
    
    RISK_CAT --> R1[Attacks USING AI]
    RISK_CAT --> R2[Attacks ON AI]
    RISK_CAT --> R3[AI Reliability<br/>Failures]
    
    R1 --> MITIGATE[Implement<br/>Mitigations]
    R2 --> MITIGATE
    R3 --> MITIGATE
    
    MITIGATE --> INCIDENT{AI Incident<br/>Occurs?}
    INCIDENT -->|Critical| REPORT12[Report within<br/>12 hours]
    INCIDENT -->|Significant| REPORT72[Report within<br/>72 hours]
    INCIDENT -->|No| MONITOR[Ongoing<br/>Monitoring]
    
    REPORT12 --> REMEDIATE[Remediate and<br/>Document]
    REPORT72 --> REMEDIATE
    
    classDef start fill:#3498db,stroke:#333
    classDef decision fill:#f1c40f,stroke:#333
    classDef action fill:#2ecc71,stroke:#333
    classDef risk fill:#e67e22,stroke:#333
    
    class START start
    class SOCI,INCIDENT decision
    class CIRMP,MITIGATE,REPORT12,REPORT72,MONITOR,REMEDIATE action
    class R1,R2,R3,RISK_CAT risk
Loading

Private Sector (Voluntary)

flowchart TD
    START[Private Sector<br/>Organisation] --> SECTOR{Regulated<br/>Sector?}
    
    SECTOR -->|Financial Services| APRA[APRA CPS 234/230<br/>ASIC Guidance]
    SECTOR -->|Online Services| ESAFE[eSafety<br/>Standards]
    SECTOR -->|Healthcare| TGA[TGA Medical<br/>Device Rules]
    SECTOR -->|General| VOL[Voluntary<br/>Frameworks Only]
    
    APRA --> ACSC[ACSC Guidance<br/>Recommended]
    ESAFE --> ACSC
    TGA --> ACSC
    VOL --> ACSC
    
    ACSC --> AI6[NAIC AI6<br/>Guidance]
    AI6 --> ETHICS[AI Ethics<br/>Principles]
    
    ETHICS --> IMPLEMENT[Implement Based<br/>on Risk Appetite]
    
    classDef start fill:#3498db,stroke:#333
    classDef decision fill:#f1c40f,stroke:#333
    classDef sector fill:#9b59b6,stroke:#333
    classDef voluntary fill:#1abc9c,stroke:#333
    
    class START start
    class SECTOR decision
    class APRA,ESAFE,TGA sector
    class VOL,ACSC,AI6,ETHICS,IMPLEMENT voluntary
Loading

Entity Relationship Model

For knowledge graph construction, use these entity types and relationships:

Entity Types

entities:
  Document:
    properties:
      - id: string            # e.g., "ISM-2072", "PSPF-ADV-001-2025"
      - title: string
      - version: string
      - date: date
      - url: string
      - status: enum[mandatory, voluntary, guidance, framework, superseded]
      - scope: enum[government, sector, national, all]
      - description: text

  Authority:
    properties:
      - id: string            # e.g., "ACSC", "APRA", "DIGITAL-NSW"
      - name: string
      - abbreviation: string
      - jurisdiction: enum[federal, nsw, vic, qld, sa, wa, tas, nt, act, national, international]
      - type: enum[department, agency, regulator, body, institute, office]
      - parent: Authority     # e.g., ACSC -> ASD
      - website: string

  Control:
    properties:
      - id: string            # e.g., "ISM-1923"
      - title: string
      - requirement: text     # Official control text
      - applicability: text
      - date_added: date
      - external_reference: string  # e.g., "OWASP Top 10 LLM v2.0"

  Sector:
    properties:
      - id: string
      - name: string
      - regulator: Authority
      - soci_designated: boolean  # Under SOCI Act?

  RiskCategory:
    properties:
      - id: string
      - name: string
      - level: enum[low, medium, high, very_high, prohibited]
      - framework: Document
      - description: text

  InternationalFramework:
    properties:
      - id: string            # e.g., "EU-AI-ACT", "NIST-AI-RMF"
      - title: string
      - jurisdiction: string  # e.g., "EU", "US", "UK", "Singapore"
      - status: enum[binding, voluntary, proposed]
      - date: date
      - url: string

  Gap:
    properties:
      - id: string            # e.g., "GAP-01"
      - title: string
      - description: text
      - impact: enum[low, medium, high, critical]
      - effort_to_close: enum[low, medium, high]
      - priority: enum[critical, high, medium, low]

  Standard:
    properties:
      - id: string            # e.g., "ISO-42001-2023", "NIST-AI-100-2"
      - title: string
      - organisation: string  # ISO, NIST, IEEE
      - publication_id: string  # Full publication number
      - status: enum[published, draft, in_development]
      - date: date
      - url: string

Relationship Types

relationships:
  ISSUED_BY:
    from: Document
    to: Authority

  REFERENCES:
    from: Document
    to: Document
    properties:
      - type: enum[mandatory, recommended, informative]

  CONTAINS:
    from: Document
    to: Control

  APPLIES_TO:
    from: Document
    to: Sector

  SUPERSEDES:
    from: Document
    to: Document
    properties:
      - date: date

  ALIGNS_WITH:
    from: Document
    to: Standard
    properties:
      - mapping_type: enum[full, partial, referenced]

  REGULATES:
    from: Authority
    to: Sector

  ADDRESSES:
    from: Document
    to: Gap
    properties:
      - coverage: enum[full, partial, none]

  EQUIVALENT_TO:
    from: Document
    to: InternationalFramework
    properties:
      - comparison: text  # How they compare

  MITIGATES:
    from: Control
    to: RiskCategory

  PARENT_OF:
    from: Authority
    to: Authority

  COORDINATES_WITH:
    from: Authority
    to: Authority
    properties:
      - mechanism: string  # e.g., "Five Eyes", "INASI", "Data and Digital Ministers"

Instance Examples

# Example: ISM-1923 control and its relationships
instances:
  - entity: Control
    id: "ISM-1923"
    title: "OWASP Top 10 for LLM"
    requirement: "Risks identified in the OWASP Top 10 for Large Language Model Applications are mitigated."
    applicability: "All LLM implementations in Commonwealth entities"
    date_added: "2024-06-01"
    external_reference: "OWASP Top 10 for LLM Applications v2.0 (2025)"

  - relationship: CONTAINS
    from: {type: Document, id: "ISM"}
    to: {type: Control, id: "ISM-1923"}

  - relationship: ISSUED_BY
    from: {type: Document, id: "ISM"}
    to: {type: Authority, id: "ACSC"}

  - relationship: ALIGNS_WITH
    from: {type: Document, id: "ISM"}
    to: {type: Standard, id: "ISO-27001-2022"}
    mapping_type: "full"

  - relationship: ADDRESSES
    from: {type: Document, id: "ISM"}
    to: {type: Gap, id: "GAP-05"}
    coverage: "partial"  # ISM-1924 addresses detection, not pre-deployment test

Sample Graph Queries

Neo4j Cypher Examples

Find all mandatory documents:

MATCH (d:Document {status: 'mandatory'})
RETURN d.title, d.date
ORDER BY d.date DESC

Find documents that reference ISM:

MATCH (d:Document)-[:REFERENCES]->(ism:Document {title: 'Information Security Manual'})
RETURN d.title

Find compliance path for Commonwealth entity:

MATCH path = (start:Document {title: 'ISM'})-[:REFERENCES*1..3]->(end:Document)<-[:ISSUED_BY]-(auth:Authority)
WHERE auth.jurisdiction = 'federal'
RETURN path

Find all AI controls:

MATCH (d:Document)-[:CONTAINS]->(c:Control)
WHERE c.title CONTAINS 'AI' OR c.title CONTAINS 'LLM'
RETURN d.title, c.id, c.title, c.requirement

Find gaps not addressed by any document:

MATCH (g:Gap)
WHERE NOT EXISTS { MATCH (d:Document)-[:ADDRESSES {coverage: 'full'}]->(g) }
RETURN g.id, g.title, g.impact, g.priority
ORDER BY g.priority

Find international equivalents for Australian frameworks:

MATCH (d:Document)-[:EQUIVALENT_TO]->(intl:InternationalFramework)
RETURN d.title AS australian_framework, intl.title AS international_equivalent, intl.jurisdiction

Find which controls mitigate which risk categories:

MATCH (c:Control)-[:MITIGATES]->(r:RiskCategory)
RETURN c.id, c.title, r.name, r.level
ORDER BY r.level DESC

Find Five Eyes coordination network:

MATCH (a1:Authority)-[coord:COORDINATES_WITH]->(a2:Authority)
WHERE coord.mechanism = 'Five Eyes'
RETURN a1.name, a2.name, coord.mechanism

Data Files

Data files for knowledge graph construction are planned for a future release. See the CHANGELOG for the roadmap.

Visualisation Tools

Recommended tools for visualising the knowledge graph:

Tool Purpose Link
Neo4j Graph database and visualisation neo4j.com
Obsidian Markdown-based knowledge graph obsidian.md
Kumu Relationship mapping kumu.io
draw.io Diagram creation diagrams.net
Mermaid Code-based diagrams mermaid.js.org

Contributing

Help improve the knowledge graph:

  1. Add relationships - Identify connections between documents
  2. Update metadata - Correct dates, versions, URLs
  3. Extend coverage - Add missing documents
  4. Build tools - Create visualisations or query interfaces

See CONTRIBUTING.md for guidelines.

Back to Index | Gap Analysis | Inventory