@cdxgen/cdxgen 0.0.1 → 13.0.1

package/data/rules/ai-bom.yaml

@@ -0,0 +1,268 @@
+- id: AIG-001
+  name: "Prompt config file ships in a build or post-build BOM"
+  description: "Prompt and model-routing config files included in distributable BOMs deserve governance review because they can silently alter downstream AI behavior, provider selection, and policy posture."
+  severity: medium
+  category: ai-governance
+  dry-run-support: full
+  standards:
+    owasp-ai-top-10:
+      - "LLM08: Excessive Agency"
+      - "LLM09: Overreliance"
+      - "LLM06:2025 Excessive Agency"
+    eu-ai-act:
+      - "Technical documentation and traceability"
+      - "Transparency and instructions for use"
+    nist-ai-rmf:
+      - "Govern"
+      - "Map"
+  condition: |
+    $count(metadata.lifecycles[phase = 'build' or phase = 'post-build']) > 0
+      ? components[
+          $prop($, 'cdx:file:kind') = 'prompt-config-file'
+        ]
+      : []
+  location: |
+    {
+      "bomRef": $."bom-ref"
+    }
+  message: "Prompt or model-routing file '{{ name }}' is included in a build/post-build BOM"
+  mitigation: "Review shipped prompt/config files explicitly, keep model-routing decisions versioned, and exclude them from release BOMs when they are only development-time inputs."
+  evidence: |
+    {
+      "fileKind": $prop($, 'cdx:file:kind'),
+      "source": $prop($, 'cdx:ai:source')
+    }
+
+- id: AIG-002
+  name: "AI inference service lacks explicit model selection"
+  description: "Source-discovered AI services without an explicit model identifier make governance, reproducibility, and downstream policy gating harder."
+  severity: medium
+  category: ai-governance
+  dry-run-support: full
+  standards:
+    owasp-ai-top-10:
+      - "LLM09: Overreliance"
+    eu-ai-act:
+      - "Technical documentation and traceability"
+      - "Human oversight"
+    nist-ai-rmf:
+      - "Govern"
+      - "Manage"
+  condition: |
+    $auditServices($)[
+      $prop($, 'cdx:ai:kind') = 'inference-service'
+      and $prop($, 'cdx:ai:modelSelection') = 'implicit'
+    ]
+  location: |
+    {
+      "bomRef": $."bom-ref",
+      "purl": endpoints[0]
+    }
+  message: "AI inference service '{{ name }}' is used without an explicit model identifier"
+  mitigation: "Declare the exact model ID in code or configuration so routing decisions can be reviewed, tested, and governed consistently."
+  evidence: |
+    {
+      "provider": $prop($, 'cdx:ai:provider'),
+      "deployment": $prop($, 'cdx:ai:deployment'),
+      "modelCount": $prop($, 'cdx:ai:modelCount'),
+      "source": $prop($, 'cdx:ai:source')
+    }
+
+- id: AIS-001
+  name: "AI inference service uses an insecure remote HTTP endpoint"
+  description: "Remote AI inference endpoints should use HTTPS so prompts, tokens, and model responses are not exposed to passive interception or tampering."
+  severity: high
+  category: ai-security
+  dry-run-support: full
+  attack:
+    tactics: [TA0001, TA0009]
+    techniques: [T1190, T1557]
+  standards:
+    owasp-ai-top-10:
+      - "LLM07: Insecure Plugin Design"
+      - "LLM05: Supply Chain Vulnerabilities"
+      - "LLM03:2025 Supply Chain"
+    eu-ai-act:
+      - "Accuracy, robustness and cybersecurity"
+      - "Technical documentation and traceability"
+    eu-cra:
+      - "Secure by default network exposure"
+      - "Technical documentation for software components"
+    nist-ai-rmf:
+      - "Manage"
+  condition: |
+    $auditServices($)[
+      $prop($, 'cdx:ai:kind') = 'inference-service'
+      and $prop($, 'cdx:ai:deployment') = 'remote'
+      and $prop($, 'cdx:ai:transportSecurity') = 'insecure-http'
+    ]
+  location: |
+    {
+      "bomRef": $."bom-ref",
+      "purl": endpoints[0]
+    }
+  message: "AI inference service '{{ name }}' uses insecure remote transport '{{ endpoints[0] }}'"
+  mitigation: "Require HTTPS/TLS for remote AI endpoints and keep any development-only HTTP endpoints isolated to local-only environments."
+  evidence: |
+    {
+      "provider": $prop($, 'cdx:ai:provider'),
+      "deployment": $prop($, 'cdx:ai:deployment'),
+      "transportSecurity": $prop($, 'cdx:ai:transportSecurity')
+    }
+
+- id: AIG-003
+  name: "AI model is a derived or modified variant"
+  description: "Fine-tuned, distilled, merged, quantized, unlocked, or otherwise modified models deserve explicit review because their behavior and deployment profile can differ materially from the upstream base model."
+  severity: medium
+  category: ai-governance
+  dry-run-support: full
+  standards:
+    owasp-ai-top-10:
+      - "LLM05: Supply Chain Vulnerabilities"
+      - "LLM03:2025 Supply Chain"
+      - "LLM09: Overreliance"
+    eu-ai-act:
+      - "Technical documentation and traceability"
+      - "Transparency and instructions for use"
+    nist-ai-rmf:
+      - "Govern"
+      - "Map"
+  condition: |
+    components[
+      type = 'machine-learning-model'
+      and $count(properties[
+        name = 'cdx:ai:variant'
+        and (
+          value = 'fine-tuned'
+          or value = 'distilled'
+          or value = 'merged'
+          or value = 'adapter'
+          or value = 'quantized'
+          or value = 'unlocked'
+          or value = 'abliterated'
+        )
+      ]) > 0
+    ]
+  location: |
+    {
+      "bomRef": $."bom-ref"
+    }
+  message: "AI model '{{ name }}' is marked as a modified variant"
+  mitigation: "Review how the model was adapted, quantized, or otherwise modified; record the upstream source model and confirm the derived artifact still meets governance, safety, and performance expectations."
+  evidence: |
+    {
+      "provider": $prop($, 'cdx:ai:provider'),
+      "source": $prop($, 'cdx:ai:source'),
+      "pedigreeNotes": pedigree.notes
+    }
+
+- id: AIS-002
+  name: "AI model appears unlocked or abliterated"
+  description: "Unlocked or abliterated models often indicate safety controls were weakened or removed, increasing misuse and policy-bypass risk."
+  severity: high
+  category: ai-security
+  dry-run-support: full
+  attack:
+    tactics: [TA0001, TA0005]
+    techniques: [T1190, T1552]
+  standards:
+    owasp-ai-top-10:
+      - "LLM06:2025 Excessive Agency"
+      - "LLM09: Overreliance"
+    eu-ai-act:
+      - "Accuracy, robustness and cybersecurity"
+      - "Technical documentation and traceability"
+    nist-ai-rmf:
+      - "Manage"
+      - "Govern"
+  condition: |
+    components[
+      type = 'machine-learning-model'
+      and $count(properties[
+        name = 'cdx:ai:variant'
+        and (value = 'unlocked' or value = 'abliterated')
+      ]) > 0
+    ]
+  location: |
+    {
+      "bomRef": $."bom-ref"
+    }
+  message: "AI model '{{ name }}' appears unlocked or abliterated"
+  mitigation: "Treat unlocked or abliterated models as high-risk artifacts, review intended use carefully, and add deployment guardrails before distribution."
+  evidence: |
+    {
+      "provider": $prop($, 'cdx:ai:provider'),
+      "source": $prop($, 'cdx:ai:source'),
+      "pedigreeNotes": pedigree.notes
+    }
+
+- id: AIP-001
+  name: "Local AI model advertises a very large context window"
+  description: "Large local context windows can increase memory pressure, latency, and review cost for self-hosted AI runtimes."
+  severity: medium
+  category: ai-performance
+  dry-run-support: full
+  standards:
+    eu-ai-act:
+      - "Performance and resource limitations should be documented"
+      - "Technical documentation and traceability"
+    nist-ai-rmf:
+      - "Measure"
+      - "Manage"
+  condition: |
+    components[
+      type = 'machine-learning-model'
+      and (
+        $contains($nullSafeProp($, 'cdx:ai:runtime'), 'ollama')
+        or $contains($nullSafeProp($, 'cdx:ai:runtime'), 'llama.cpp')
+      )
+      and $number($nullSafeProp($, 'cdx:ai:contextWindow')) >= 131072
+    ]
+  location: |
+    {
+      "bomRef": $."bom-ref"
+    }
+  message: "Local AI model '{{ name }}' advertises a large context window of {{ $prop($, 'cdx:ai:contextWindow') }} tokens"
+  mitigation: "Validate that the target environment has enough memory and latency budget for the configured context window, or choose a smaller context budget."
+  evidence: |
+    {
+      "runtime": $prop($, 'cdx:ai:runtime'),
+      "contextWindow": $prop($, 'cdx:ai:contextWindow'),
+      "artifactFormat": $prop($, 'cdx:ai:artifactFormat')
+    }
+
+- id: AIP-002
+  name: "Large local AI model lacks quantization metadata"
+  description: "Large locally hosted models without quantization metadata are harder to capacity-plan and can hide unnecessary CPU, RAM, or GPU pressure."
+  severity: medium
+  category: ai-performance
+  dry-run-support: full
+  standards:
+    eu-ai-act:
+      - "Technical documentation and traceability"
+      - "Performance and deployment characteristics should be documented"
+    nist-ai-rmf:
+      - "Measure"
+      - "Manage"
+  condition: |
+    components[
+      type = 'machine-learning-model'
+      and (
+        $prop($, 'cdx:ai:source') = 'local-artifact'
+        or $prop($, 'cdx:ai:source') = 'local-config'
+      )
+      and $number($nullSafeProp($, 'cdx:ai:parameterCount')) >= 7000000000
+      and $nullSafeProp($, 'cdx:ai:quantization') = ''
+    ]
+  location: |
+    {
+      "bomRef": $."bom-ref"
+    }
+  message: "Large local AI model '{{ name }}' lacks quantization metadata"
+  mitigation: "Record quantization and deployment sizing details for large local models so reviewers can validate resource budgets and runtime fit before rollout."
+  evidence: |
+    {
+      "runtime": $prop($, 'cdx:ai:runtime'),
+      "parameterCount": $prop($, 'cdx:ai:parameterCount'),
+      "source": $prop($, 'cdx:ai:source')
+    }

package/data/rules/ai-oversight.yaml

@@ -0,0 +1,186 @@
+# AI Human Oversight & Rigor rules.
+#
+# These rules evaluate the cdx:ai:oversight:* properties (BOM document root) emitted by the
+# AI oversight collector. They evaluate whether AI-assisted code was merged
+# with adequate independent human review, and detect rubber-stamping or quality gate bypassing.
+
+- id: AIOVS-001
+  name: "High AI Verification Debt"
+  description: >-
+    A high fraction of AI-authored commits or pull requests lacks independent
+    human review, suggesting a potential failure of human-in-the-loop controls.
+  severity: high
+  category: ai-oversight
+  dry-run-support: full
+  standards:
+    eu-ai-act:
+      - "Article 14 - Human oversight"
+    nist-ai-rmf:
+      - "Govern"
+      - "Manage"
+  condition: |
+    $[$number($nullSafeProp($, 'cdx:ai:oversight:verificationDebtRatio')[$ != '' and $ != 'unavailable']) >= 0.5]
+  location: |
+    {
+      "purl": metadata.component.purl,
+      "bomRef": metadata.component."bom-ref"
+    }
+  message: "High AI verification debt detected ({{ $prop(bom, 'cdx:ai:oversight:verificationDebtRatio') }}). A large fraction of AI-generated code has not been independently reviewed."
+  mitigation: >-
+    Ensure all AI-generated code undergoes independent human review and approval before merging.
+    Establish pull request guidelines that require at least one human reviewer to sign off.
+  evidence: |
+    {
+      "verificationDebtRatio": $prop(bom, 'cdx:ai:oversight:verificationDebtRatio'),
+      "selfMergeRate": $prop(bom, 'cdx:ai:oversight:selfMergeRate'),
+      "score": $prop(bom, 'cdx:ai:oversight:score')
+    }
+
+- id: AIOVS-002
+  name: "CI Weakening Events in AI Commits"
+  description: >-
+    AI-authored commits were found to weaken quality gates by deleting test files,
+    adding silencing scripts (like '|| true' or 'continue-on-error'), or skipping lints.
+  severity: high
+  category: ai-oversight
+  dry-run-support: full
+  standards:
+    eu-ai-act:
+      - "Article 15 - Accuracy, robustness and cybersecurity"
+    nist-ai-rmf:
+      - "Measure"
+      - "Manage"
+  condition: |
+    $[$number($nullSafeProp($, 'cdx:ai:oversight:ciWeakeningEvents')[$ != '']) > 0]
+  location: |
+    {
+      "purl": metadata.component.purl,
+      "bomRef": metadata.component."bom-ref"
+    }
+  message: "CI weakening events detected in AI commits ({{ $prop(bom, 'cdx:ai:oversight:ciWeakeningEvents') }} event(s))."
+  mitigation: >-
+    Review AI commits that modified CI/CD configurations, bypassed tests, or deleted checks.
+    Re-enable tests and ensure quality gates cannot be bypassed by automated changes.
+  evidence: |
+    {
+      "ciWeakeningEvents": $prop(bom, 'cdx:ai:oversight:ciWeakeningEvents'),
+      "score": $prop(bom, 'cdx:ai:oversight:score')
+    }
+
+- id: AIOVS-003
+  name: "High Self-Merge Rate of AI Commits"
+  description: >-
+    A large fraction of AI-authored commits were self-merged or self-approved by
+    their author, bypassing independent validation and peer review.
+  severity: medium
+  category: ai-oversight
+  dry-run-support: full
+  standards:
+    eu-ai-act:
+      - "Article 14 - Human oversight"
+    nist-ai-rmf:
+      - "Govern"
+      - "Measure"
+  condition: |
+    $[$number($nullSafeProp($, 'cdx:ai:oversight:selfMergeRate')[$ != '' and $ != 'unavailable']) >= 0.5]
+  location: |
+    {
+      "purl": metadata.component.purl,
+      "bomRef": metadata.component."bom-ref"
+    }
+  message: "High self-merge rate of AI commits detected ({{ $prop(bom, 'cdx:ai:oversight:selfMergeRate') }})."
+  mitigation: >-
+    Require independent review for all pull requests. Configure repository branch protection
+    rules to prevent self-merging and require independent approvals.
+  evidence: |
+    {
+      "selfMergeRate": $prop(bom, 'cdx:ai:oversight:selfMergeRate'),
+      "score": $prop(bom, 'cdx:ai:oversight:score')
+    }
+
+- id: AIOVS-004
+  name: "Low AI Review Coverage"
+  description: >-
+    The percentage of AI commits/PRs with verified independent human approvals is low.
+  severity: medium
+  category: ai-oversight
+  dry-run-support: full
+  standards:
+    eu-ai-act:
+      - "Article 14 - Human oversight"
+    nist-ai-rmf:
+      - "Govern"
+  condition: |
+    $[$number($nullSafeProp($, 'cdx:ai:oversight:reviewCoverage')[$ != '' and $ != 'unavailable']) < 0.5]
+  location: |
+    {
+      "purl": metadata.component.purl,
+      "bomRef": metadata.component."bom-ref"
+    }
+  message: "Low AI review coverage detected ({{ $prop(bom, 'cdx:ai:oversight:reviewCoverage') }})."
+  mitigation: >-
+    Require Pull Request reviews for all branches and increase reviewer coverage to ensure independent validation.
+  evidence: |
+    {
+      "reviewCoverage": $prop(bom, 'cdx:ai:oversight:reviewCoverage'),
+      "score": $prop(bom, 'cdx:ai:oversight:score')
+    }
+
+- id: AIOVS-005
+  name: "Weak Role Accountability and Gaps in AI Attribution"
+  description: >-
+    A large fraction of files touched by AI-authored commits are not covered by a
+    CODEOWNERS entry, indicating weak role accountability and gaps in AI attribution.
+    Fires only when a CODEOWNERS file exists (otherwise coverage is `unavailable`).
+  severity: low
+  category: ai-oversight
+  dry-run-support: full
+  standards:
+    eu-ai-act:
+      - "Article 12 - Record-keeping"
+    nist-ai-rmf:
+      - "Govern"
+  condition: |
+    $[$number($nullSafeProp($, 'cdx:ai:oversight:codeownersCoverage')[$ != '' and $ != 'unavailable']) < 0.5]
+  location: |
+    {
+      "purl": metadata.component.purl,
+      "bomRef": metadata.component."bom-ref"
+    }
+  message: "Weak role accountability or sign-off on AI commits: CODEOWNERS ({{ $prop(bom, 'cdx:ai:oversight:codeownersCoverage') }}), Sign-off ({{ $prop(bom, 'cdx:ai:oversight:signoffCoverage') }})."
+  mitigation: >-
+    Define file ownership in CODEOWNERS and enforce developer sign-offs (DCO) to improve accountability.
+  evidence: |
+    {
+      "codeownersCoverage": $prop(bom, 'cdx:ai:oversight:codeownersCoverage'),
+      "signoffCoverage": $prop(bom, 'cdx:ai:oversight:signoffCoverage')
+    }
+
+- id: AIOVS-006
+  name: "Weak Overall AI Human Oversight"
+  description: >-
+    The overall human oversight score falls into the weak band, indicating poor governance.
+  severity: medium
+  category: ai-oversight
+  dry-run-support: full
+  standards:
+    eu-ai-act:
+      - "Article 14 - Human oversight"
+    nist-ai-rmf:
+      - "Govern"
+      - "Manage"
+  condition: |
+    $[$nullSafeProp($, 'cdx:ai:oversight:band') = 'weak']
+  location: |
+    {
+      "purl": metadata.component.purl,
+      "bomRef": metadata.component."bom-ref"
+    }
+  message: "Weak overall AI human oversight score ({{ $prop(bom, 'cdx:ai:oversight:score') }})."
+  mitigation: >-
+    Address high-priority governance gaps including verification debt, self-merges, and CI weakening to establish a moderate or strong oversight band.
+  evidence: |
+    {
+      "score": $prop(bom, 'cdx:ai:oversight:score'),
+      "band": $prop(bom, 'cdx:ai:oversight:band')
+    }

package/data/rules/ai-provenance.yaml

@@ -0,0 +1,92 @@
+# AI-generated code and authorship provenance rules.
+#
+# These rules evaluate the cdx:ai:codegen:* properties (BOM document root) emitted by the
+# AI provenance collector (see docs/AI_PROVENANCE.md). They surface governance,
+# licensing, and intellectual-property considerations that arise when AI, LLMs,
+# or coding agents contribute to a project.
+
+- id: AIPROV-001
+  name: "AI/LLM/agent involvement detected in project development"
+  description: >-
+    Signals of AI, LLM, or coding-agent involvement were detected in this
+    project's development, testing, review, or triage. Under current US Copyright
+    Office guidance (Copyright and Artificial Intelligence, Part 2, 2025) and
+    Thaler v. Perlmutter, works lacking sufficient human authorship are not
+    eligible for copyright protection - prompts alone do not make the user the
+    author of the output. Code produced solely by AI may therefore receive no
+    implicit copyright protection and could be freely used by others unless it is
+    protected by another mechanism such as a licence, contract, or trade secret.
+  severity: low
+  category: ai-provenance
+  dry-run-support: full
+  standards:
+    eu-ai-act:
+      - "Transparency and instructions for use"
+      - "Technical documentation and traceability"
+    nist-ai-rmf:
+      - "Govern"
+      - "Map"
+  condition: |
+    $[$nullSafeProp($, 'cdx:ai:codegen:detected') = 'true']
+  location: |
+    {
+      "purl": metadata.component.purl,
+      "bomRef": metadata.component."bom-ref"
+    }
+  message: "AI/LLM/agent involvement detected ({{ $prop(bom, 'cdx:ai:codegen:tools') }}); AI-authored code may receive no implicit copyright protection without sufficient human authorship."
+  mitigation: >-
+    Document the creative human contributions across the software development
+    lifecycle (design, selection, arrangement, and substantial modification of
+    AI output). Confirm the licensing and ownership posture for AI-assisted code
+    and consider contractual or trade-secret protections where copyright may not
+    apply. Attribute AI usage in commits and disclosures for auditability.
+  evidence: |
+    {
+      "tools": $prop(bom, 'cdx:ai:codegen:tools'),
+      "phases": $prop(bom, 'cdx:ai:codegen:phases'),
+      "confidence": $prop(bom, 'cdx:ai:codegen:confidence'),
+      "confidenceBand": $prop(bom, 'cdx:ai:codegen:confidence:band'),
+      "signalCount": $prop(bom, 'cdx:ai:codegen:signals:count')
+    }
+
+- id: AIPROV-002
+  name: "High-confidence AI/LLM/agent authorship signals"
+  description: >-
+    High-confidence signals (for example bot commit authors or explicit
+    AI co-author trailers) indicate that AI or coding agents authored parts of
+    this project. Purely AI-generated code is ineligible for copyright protection
+    in the absence of sufficiently creative human input, which has ownership,
+    licensing, and warranty implications for downstream consumers.
+  severity: medium
+  category: ai-provenance
+  dry-run-support: full
+  standards:
+    eu-ai-act:
+      - "Transparency and instructions for use"
+      - "Human oversight"
+    nist-ai-rmf:
+      - "Govern"
+      - "Manage"
+  condition: |
+    $[
+      $nullSafeProp($, 'cdx:ai:codegen:detected') = 'true'
+      and $nullSafeProp($, 'cdx:ai:codegen:confidence:band') = 'high'
+    ]
+  location: |
+    {
+      "purl": metadata.component.purl,
+      "bomRef": metadata.component."bom-ref"
+    }
+  message: "High-confidence AI/LLM/agent authorship detected ({{ $prop(bom, 'cdx:ai:codegen:tools') }}); review copyright, licensing, and human-authorship posture before relying on ownership of this code."
+  mitigation: >-
+    Treat this code as potentially uncopyrightable unless human authorship can be
+    demonstrated. Record who reviewed, modified, and arranged the AI output;
+    verify third-party licence compatibility; and communicate the AI-authorship
+    status to consumers who may assume standard copyright ownership.
+  evidence: |
+    {
+      "tools": $prop(bom, 'cdx:ai:codegen:tools'),
+      "phases": $prop(bom, 'cdx:ai:codegen:phases'),
+      "confidence": $prop(bom, 'cdx:ai:codegen:confidence'),
+      "signalCount": $prop(bom, 'cdx:ai:codegen:signals:count')
+    }