Skip to content

FRAMEWORK_MAPPINGS.md

Latest commit

 

History

History
180 lines (141 loc) · 9.3 KB

FRAMEWORK_MAPPINGS.md

File metadata and controls

180 lines (141 loc) · 9.3 KB

Framework Mappings

This document shows which frameworks are represented in cloud-ai-security-skills, where they appear today, and where coverage is still partial.

For machine-readable source of truth, see docs/framework-coverage.json. For the policy behind that file, see docs/COVERAGE_MODEL.md.

The repo uses two mapping styles:

  • Event / finding mappings inside OCSF output, especially finding_info.attacks[]
  • Skill-level control mappings inside benchmark and evaluation skills

The goal is to make coverage explicit instead of forcing reviewers to infer it from individual SKILL.md files.

Current posture

Framework Status Where it appears
OCSF 1.8 core wire contract all ingestion, detection, and view flows
MITRE ATT&CK v14 strong cloud, Kubernetes, container, MCP, and remediation skills
MITRE ATLAS partial but real AI-oriented evaluation and discovery skills
CIS Benchmarks / Controls strong AWS, GCP, Azure, Kubernetes, container evaluation skills
NIST CSF 2.0 strong evaluation and some remediation skills
OWASP LLM Top 10 partial model-serving and future AI inventory/enrichment paths
OWASP MCP Top 10 partial detect-mcp-tool-drift and MCP-related repo controls
SOC 2 TSC partial evaluation and remediation mappings
ISO 27001:2022 partial CSPM/evaluation mappings
PCI DSS 4.0 partial AWS posture mappings today
NIST AI RMF real and growing model-serving security, cloud control evidence, AI BOM, and roadmap items

By layer

Layer Main frameworks
ingestion/ OCSF 1.8, vendor schemas, source-specific event contracts
discovery/ MITRE ATT&CK, MITRE ATLAS, CycloneDX ML-BOM, NIST AI RMF, PCI / SOC 2 evidence support
detection/ OCSF 1.8, MITRE ATT&CK, selective OWASP MCP mappings
evaluation/ CIS, NIST CSF, ISO, SOC 2, PCI, MITRE ATLAS, OWASP LLM
view/ OCSF 1.8, SARIF, Mermaid, MITRE ATT&CK labels
remediation/ MITRE ATT&CK, NIST CSF, CIS Controls, SOC 2

MITRE ATT&CK

Strongest current ATT&CK coverage:

Skill Coverage
detect-lateral-movement T1021, T1078.004 across AWS role sessions, GCP service-account pivots, Azure Activity role/managed-identity pivots, and Azure Entra / Graph application-service-principal credential pivots
detect-okta-mfa-fatigue T1621 for repeated Okta Verify push challenge + deny bursts
detect-entra-credential-addition T1098.001 for successful Entra application or service-principal credential additions and federated identity credential creation
detect-entra-role-grant-escalation T1098.003 for successful Entra app-role assignments that grant additional application permissions to service principals
detect-google-workspace-suspicious-login T1110 and T1078 for provider-marked Google Workspace suspicious logins and repeated failure-then-success bursts
detect-mcp-tool-drift T1195.001
detect-privilege-escalation-k8s T1552.007, T1611, T1098, T1550.001
detect-sensitive-secret-read-k8s T1552, T1552.007
ingest-guardduty-ocsf curated ATT&CK tactic/technique extraction
ingest-security-hub-ocsf ATT&CK extraction when upstream findings contain MITRE hints
iam-departures-remediation ATT&CK-linked remediation context
discover-environment ATT&CK graph overlay for cloud resources and relationships

Notes:

  • ATT&CK is pinned in the shared OCSF contract.
  • ATT&CK mappings belong inside finding_info.attacks[] for OCSF 1.8 outputs, not as loose side metadata.
  • Current cross-cloud identity depth is strongest in detect-lateral-movement; the Azure slice now distinguishes Azure Activity control-plane pivots from Entra / Graph application-service-principal credential pivots.

OCSF identity normalization

Identity ingestion is broader than cloud control planes now.

Skill OCSF scope
ingest-entra-directory-audit-ocsf API Activity (6003) for verified Microsoft Entra / Graph directoryAudit identity-management events
ingest-google-workspace-login-ocsf Authentication (3002) and Account Change (3001) for verified Google Workspace Admin SDK Reports login-audit events
ingest-okta-system-log-ocsf Authentication (3002), Account Change (3001), User Access Management (3005) from verified Okta System Log fields, including Okta Verify push and deny event families
ingest-cloudtrail-ocsf API Activity (6003) for AWS IAM and control-plane events
ingest-gcp-audit-ocsf API Activity (6003) for GCP audit events
ingest-azure-activity-ocsf API Activity (6003) for Azure Activity events

The rule stays the same across vendors:

  • verify the real source payload and natural IDs first
  • normalize to the narrowest OCSF class that fits cleanly
  • keep provider-specific context under unmapped instead of losing it
  • keep source-specific event families separate instead of pretending Okta, Entra, and Workspace login telemetry are interchangeable

Current identity-vendor detections on top of those streams:

  • detect-okta-mfa-fatigue
  • detect-entra-credential-addition
  • detect-entra-role-grant-escalation
  • detect-google-workspace-suspicious-login

MITRE ATLAS

ATLAS is present today, but coverage is narrower than ATT&CK.

Skill Coverage
discover-environment graph overlay for AI/ML resources and adversarial ML techniques
model-serving-security explicit ATLAS coverage plus machine-readable NIST AI RMF section scope in the benchmark metadata
discover-ai-bom inventory artifact for future ATLAS / AI RMF evidence joins
discover-control-evidence evidence package that preserves ATLAS / AI RMF context from discovery artifacts
discover-cloud-control-evidence cross-cloud evidence package with explicit NIST AI RMF evidence mode and ATT&CK / ATLAS / AI RMF inventory context

Current provider depth in discovery:

  • AWS Bedrock / SageMaker
  • Google Vertex AI
  • Azure ML / Azure AI Foundry

Recommended expansion:

  • add a shared table of ATLAS techniques used by AI-oriented skills
  • deepen provider-shaped GPU and accelerator evidence beyond the current benchmark input model

Kubernetes and containers

Kubernetes and container security are currently covered through both benchmark and threat-model mappings.

Domain Frameworks Current skills
Kubernetes detection MITRE ATT&CK detect-privilege-escalation-k8s, detect-sensitive-secret-read-k8s
Kubernetes posture CIS Kubernetes, NIST CSF k8s-security-benchmark
Container posture CIS Docker, NIST CSF container-security
Container / workload findings MITRE ATT&CK where applicable detection skills and upstream ingestors

GPU and AI infrastructure

AI and GPU security are already in scope, but their framework coverage is less uniform than classic cloud posture.

Skill Frameworks called out today
discover-ai-bom CycloneDX ML-BOM, NIST AI RMF, MITRE ATLAS, PCI, SOC 2
discover-control-evidence PCI DSS 4.0, SOC 2 TSC, CycloneDX ML-BOM, MITRE ATLAS
discover-cloud-control-evidence PCI DSS 4.0, SOC 2 TSC, NIST AI RMF, MITRE ATT&CK, MITRE ATLAS
model-serving-security MITRE ATLAS, NIST CSF, NIST AI RMF, OWASP LLM Top 10, SOC 2, provider-specific AI endpoint controls
gpu-cluster-security MITRE ATT&CK, MITRE ATLAS, NIST CSF, NIST AI RMF, CIS Controls, CIS Kubernetes
discover-environment MITRE ATT&CK, MITRE ATLAS, NIST CSF

Recommended next step:

  • make ATLAS coverage first-class for GPU and model-serving paths
  • connect AI BOM and discovery outputs more directly to ATLAS and AI RMF evidence views

Compliance and control frameworks

Framework Skills with explicit mappings today
CIS AWS Foundations cspm-aws-cis-benchmark
CIS GCP Foundations cspm-gcp-cis-benchmark
CIS Azure Foundations cspm-azure-cis-benchmark
CIS Kubernetes k8s-security-benchmark, gpu-cluster-security
CIS Docker container-security
NIST CSF 2.0 CSPM skills, K8s/container evaluation, AI-oriented evaluation, remediation
SOC 2 TSC AWS evaluation, model-serving, IAM departures remediation
ISO 27001:2022 AWS/GCP/Azure evaluation
PCI DSS 4.0 AWS evaluation
NIST AI RMF model-serving docs and roadmap expansion

AI BOM

AI BOM is not the identity of the repo, but it is now a shipped discovery capability.

Current fit:

  • collection / inventory input in discovery/
  • normalization into a deterministic CycloneDX-aligned AI BOM artifact
  • technical evidence output via discovery-layer evidence packaging for PCI and SOC 2 reviews
  • future evaluation joins against ATLAS, NIST AI RMF, OWASP LLM Top 10, PCI, and SOC 2 evidence pipelines

That keeps AI BOM as one important capability inside a broader cloud + AI security skills repo.

How to extend mappings safely

When adding or changing framework mappings:

  1. update the skill’s SKILL.md
  2. update REFERENCES.md with only official framework or vendor sources
  3. add or update tests when mapping logic affects output
  4. update this document when the change is material
  5. keep ATT&CK / ATLAS / OCSF version pinning consistent with the shared contract