genai-security-crosswalk 2.0.0

package/data/entries/DSGAI10.json

@@ -0,0 +1,627 @@
+{
+  "id": "DSGAI10",
+  "name": "Synthetic Data and Anonymization Pitfalls",
+  "source_list": "DSGAI-2026",
+  "version": "2026-Q1",
+  "severity": "Medium",
+  "aivss_score": null,
+  "audience": [
+    "red-teamer",
+    "security-engineer",
+    "ciso",
+    "compliance",
+    "ml-engineer",
+    "ot-engineer",
+    "auditor",
+    "developer",
+    "data-engineer"
+  ],
+  "mappings": [
+    {
+      "framework": "MITRE ATLAS",
+      "control_id": "AML.T0024.000",
+      "control_name": "Membership Inference",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "Adversary probes synthetic dataset to determine whether specific individuals are re-identifiable"
+    },
+    {
+      "framework": "MITRE ATLAS",
+      "control_id": "AML.T0025",
+      "control_name": "Exfiltrate via Cyber Means",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "Re-identified individuals from synthetic data extracted via standard exfiltration paths"
+    },
+    {
+      "framework": "MITRE ATLAS",
+      "control_id": "AML.T0035",
+      "control_name": "Exfiltrate via ML Inference API",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "Synthetic training data membership confirmed and specific records reconstructed through inference API"
+    },
+    {
+      "framework": "NIST AI RMF 1.0",
+      "control_id": "GV-1.6",
+      "control_name": "Policies for data privacy",
+      "tier": "Hardening",
+      "scope": "Build",
+      "notes": "Anonymisation governance policy — legal standard, not technical checkbox"
+    },
+    {
+      "framework": "NIST AI RMF 1.0",
+      "control_id": "MS-2.6",
+      "control_name": "Testing — data leakage",
+      "tier": "Hardening",
+      "scope": "Build",
+      "notes": "Re-identification risk testing and membership inference testing on synthetic datasets"
+    },
+    {
+      "framework": "NIST AI RMF 1.0",
+      "control_id": "MS-3.3",
+      "control_name": "Data quality",
+      "tier": "Hardening",
+      "scope": "Build",
+      "notes": "Data quality measurement of synthetic datasets — statistical fidelity vs privacy tradeoffs"
+    },
+    {
+      "framework": "NIST AI RMF 1.0",
+      "control_id": "MG-3.2",
+      "control_name": "Residual risk",
+      "tier": "Hardening",
+      "scope": "Build",
+      "notes": "Residual re-identification risk documented and accepted formally before dataset release or use"
+    },
+    {
+      "framework": "EU AI Act",
+      "control_id": "Training data must be subject to appropriate data governance — privacy measures must be effective",
+      "control_name": "Art. 10 — Data and data governance",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "Anonymisation effectiveness is an Art. 10 data governance requirement, not a self-certification"
+    },
+    {
+      "framework": "EU AI Act",
+      "control_id": "GPAI training data governance documented — privacy measures included",
+      "control_name": "Art. 53(1)(a) — GPAI documentation",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "Synthetic data generation methodology and re-identification risk assessment documented for GPAI"
+    },
+    {
+      "framework": "ISO/IEC 27001:2022",
+      "control_id": "A.5.34",
+      "control_name": "Privacy and PII protection",
+      "tier": "Hardening",
+      "scope": "Build",
+      "notes": "Anonymisation must meet the standard required by applicable privacy law — not just technical anonymisation"
+    },
+    {
+      "framework": "ISO/IEC 27001:2022",
+      "control_id": "A.8.11",
+      "control_name": "Data masking",
+      "tier": "Hardening",
+      "scope": "Build",
+      "notes": "Technical anonymisation and pseudonymisation controls applied to synthetic data generation"
+    },
+    {
+      "framework": "ISO/IEC 27001:2022",
+      "control_id": "A.5.12",
+      "control_name": "Classification of information",
+      "tier": "Hardening",
+      "scope": "Build",
+      "notes": "Synthetic datasets classified based on re-identification risk, not assumed to be non-personal"
+    },
+    {
+      "framework": "ISO/IEC 27001:2022",
+      "control_id": "A.8.33",
+      "control_name": "Test information",
+      "tier": "Hardening",
+      "scope": "Build",
+      "notes": "Appropriate protection of test and synthetic data used in AI development"
+    },
+    {
+      "framework": "ISO/IEC 42001:2023",
+      "control_id": "Data — acquisition",
+      "control_name": "A.7.2",
+      "tier": "Hardening",
+      "scope": "Build",
+      "notes": "Hardening"
+    },
+    {
+      "framework": "ISO/IEC 42001:2023",
+      "control_id": "Impact assessment",
+      "control_name": "A.5.2",
+      "tier": "Hardening",
+      "scope": "Build",
+      "notes": "Hardening"
+    },
+    {
+      "framework": "ISO/IEC 42001:2023",
+      "control_id": "Data — preparation",
+      "control_name": "A.7.3",
+      "tier": "Hardening",
+      "scope": "Build",
+      "notes": "Hardening"
+    },
+    {
+      "framework": "ISO/IEC 42001:2023",
+      "control_id": "Planning — risk",
+      "control_name": "Cl.6.1",
+      "tier": "Hardening",
+      "scope": "Build",
+      "notes": "Hardening"
+    },
+    {
+      "framework": "CIS Controls v8.1",
+      "control_id": "CIS 3",
+      "control_name": "3.7 — Establish data classification scheme",
+      "tier": "Hardening",
+      "scope": "Build"
+    },
+    {
+      "framework": "CIS Controls v8.1",
+      "control_id": "CIS 18",
+      "control_name": "18.3 — Remediate penetration test findings",
+      "tier": "Hardening",
+      "scope": "Build"
+    },
+    {
+      "framework": "OWASP ASVS 4.0.3",
+      "control_id": "V8 Data Protection",
+      "control_name": "V8.3.4 — Sensitive data classified",
+      "tier": "Hardening",
+      "scope": "Build"
+    },
+    {
+      "framework": "OWASP ASVS 4.0.3",
+      "control_id": "V5 Validation",
+      "control_name": "V5.2.6 — Defined output structure",
+      "tier": "Hardening",
+      "scope": "Build"
+    },
+    {
+      "framework": "ISA/IEC 62443",
+      "control_id": "SR 4.1",
+      "control_name": "Data confidentiality",
+      "tier": "Foundational",
+      "scope": "Both",
+      "notes": "Synthetic OT datasets not automatically excluded from protection — assessed before classification change"
+    },
+    {
+      "framework": "ISA/IEC 62443",
+      "control_id": "SR 3.3",
+      "control_name": "Software and information integrity",
+      "tier": "Foundational",
+      "scope": "Both",
+      "notes": "Synthetic data generation pipeline validated — source OT data integrity maintained"
+    },
+    {
+      "framework": "ISA/IEC 62443",
+      "control_id": "Security management",
+      "control_name": "62443-2-1",
+      "tier": "Foundational",
+      "scope": "Both",
+      "notes": "OT synthetic data governance policy — when synthetic data removes OT classification obligation"
+    },
+    {
+      "framework": "NIST SP 800-82 Rev 3",
+      "control_id": "ICS vulnerabilities",
+      "control_name": "§5.3",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "In-context manipulation of OT decision support"
+    },
+    {
+      "framework": "NIST SP 800-82 Rev 3",
+      "control_id": "Risk assessment",
+      "control_name": "§6.2",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "Context poisoning risk for OT data feeds"
+    },
+    {
+      "framework": "NIST CSF 2.0",
+      "control_id": "GV.RM-06",
+      "control_name": "Risk Management Strategy",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "Risk tolerance defined for re-identification risk in synthetic datasets — legal standard, not technical checkbox"
+    },
+    {
+      "framework": "NIST CSF 2.0",
+      "control_id": "ID.RA-01",
+      "control_name": "Risk Assessment",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "Re-identification risk assessed for all synthetic datasets before use or distribution"
+    },
+    {
+      "framework": "NIST CSF 2.0",
+      "control_id": "PR.DS-01",
+      "control_name": "Data Security",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "Synthetic datasets not automatically excluded from data protection — protected until re-identification risk formally assessed"
+    },
+    {
+      "framework": "NIST CSF 2.0",
+      "control_id": "DE.CM-09",
+      "control_name": "Continuous Monitoring",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "Monitoring for re-identification attempts against synthetic datasets"
+    },
+    {
+      "framework": "SOC 2",
+      "control_id": "Synthetic datasets that are not truly anonymous subject to same retention obligations as source personal data",
+      "control_name": "P4.2 — Retention of personal information",
+      "tier": "Hardening",
+      "scope": "Both"
+    },
+    {
+      "framework": "SOC 2",
+      "control_id": "Synthetic data use must respect underlying privacy commitments if re-identification risk exists",
+      "control_name": "P5.1 — Personal information use",
+      "tier": "Hardening",
+      "scope": "Both"
+    },
+    {
+      "framework": "SOC 2",
+      "control_id": "Re-identification risk in synthetic datasets identified in risk assessment",
+      "control_name": "CC3.2 — Risk assessment",
+      "tier": "Hardening",
+      "scope": "Both"
+    },
+    {
+      "framework": "SOC 2",
+      "control_id": "Synthetic OT data and business data protected at source classification level until re-identification risk formally assessed",
+      "control_name": "C2.1 — Confidential information protection",
+      "tier": "Hardening",
+      "scope": "Both"
+    },
+    {
+      "framework": "PCI DSS v4.0",
+      "control_id": "Req 3.1.1",
+      "control_name": "Account data inventory",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "Synthetic payment datasets that may be re-identifiable included in CHD inventory"
+    },
+    {
+      "framework": "PCI DSS v4.0",
+      "control_id": "Req 3.3.1",
+      "control_name": "SAD prohibition",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "Synthetic data generated from SAD retains SAD classification — generation does not remove the prohibition"
+    },
+    {
+      "framework": "PCI DSS v4.0",
+      "control_id": "Req 3.4.1",
+      "control_name": "PAN rendering",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "Synthetic PANs that pass Luhn check treated as CHD — format-preserving synthetic PANs are in PCI scope"
+    },
+    {
+      "framework": "PCI DSS v4.0",
+      "control_id": "Req 12.3.2",
+      "control_name": "Targeted risk analysis",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "Re-identification risk in synthetic payment datasets documented in targeted risk analysis"
+    },
+    {
+      "framework": "ENISA Multilayer Framework",
+      "control_id": "L2",
+      "control_name": "Data and Model Security (DMS)",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "Synthetic data generation validated against re-identification risk — formal anonymisation assessments documented as DMS evidence"
+    },
+    {
+      "framework": "ENISA Multilayer Framework",
+      "control_id": "L2",
+      "control_name": "Governance and Risk (GOV)",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "Re-identification risk included in AI privacy impact assessment — treatment controls and residual risk accepted by data protection officer"
+    },
+    {
+      "framework": "ENISA Multilayer Framework",
+      "control_id": "L2",
+      "control_name": "Monitoring and Detection (MON)",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "Quality and privacy metrics monitored for all synthetic datasets — drift in re-identification risk triggers re-assessment"
+    },
+    {
+      "framework": "ENISA Multilayer Framework",
+      "control_id": "L1",
+      "control_name": "General ICT — Data Protection",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "Anonymisation requirements established at design time — not applied as an afterthought to an already-designed pipeline"
+    },
+    {
+      "framework": "OWASP SAMM v2.0",
+      "control_id": "D-TA",
+      "control_name": "Design / Threat Assessment",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "Map all sources that contribute to context window; assess trust level per source"
+    },
+    {
+      "framework": "OWASP SAMM v2.0",
+      "control_id": "I-SB",
+      "control_name": "Implementation / Secure Build",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "Validate and sanitise all content before context window assembly"
+    },
+    {
+      "framework": "OWASP SAMM v2.0",
+      "control_id": "V-ST",
+      "control_name": "Verification / Security Testing",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "Adversarial tests injecting malicious content via each context source"
+    },
+    {
+      "framework": "OWASP SAMM v2.0",
+      "control_id": "O-IM",
+      "control_name": "Operations / Incident Management",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "Alert on reasoning deviations correlated with external content in context"
+    },
+    {
+      "framework": "CWE/CVE",
+      "control_id": "CWE-359",
+      "control_name": "CWE-359",
+      "tier": "Hardening",
+      "scope": "Build",
+      "url": "https://cwe.mitre.org/data/definitions/359.html"
+    },
+    {
+      "framework": "CWE/CVE",
+      "control_id": "CWE-330",
+      "control_name": "CWE-330",
+      "tier": "Hardening",
+      "scope": "Build",
+      "url": "https://cwe.mitre.org/data/definitions/330.html"
+    },
+    {
+      "framework": "MAESTRO",
+      "control_id": "L2",
+      "control_name": "Data Operations",
+      "tier": "Hardening",
+      "scope": "Both"
+    },
+    {
+      "framework": "MAESTRO",
+      "control_id": "L1",
+      "control_name": "Foundation Models",
+      "tier": "Hardening",
+      "scope": "Both"
+    },
+    {
+      "framework": "AIUC-1",
+      "control_id": "B001",
+      "control_name": "Third-party adversarial robustness testing",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "Foundational"
+    },
+    {
+      "framework": "AIUC-1",
+      "control_id": "B002",
+      "control_name": "Detect adversarial input",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "Hardening"
+    },
+    {
+      "framework": "AIUC-1",
+      "control_id": "B005",
+      "control_name": "Implement real-time input filtering",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "Foundational"
+    },
+    {
+      "framework": "OWASP NHI Top 10",
+      "control_id": "Data feed service accounts with access to sensitive data that should not enter context",
+      "control_name": "NHI-5 Over-Privileged NHI",
+      "tier": "Hardening",
+      "scope": "Both",
+      "notes": "Scope data feed credentials to approved data only"
+    },
+    {
+      "framework": "NIST SP 800-218A",
+      "control_id": "PW.7.2-PS",
+      "control_name": "Review for security vulnerabilities — synthetic data quality review",
+      "tier": "Foundational",
+      "scope": "Build",
+      "notes": "Review synthetic data for bias inheritance, privacy leakage, and statistical fidelity before use in training pipelines; verify privacy guarantees are meaningful"
+    },
+    {
+      "framework": "NIST SP 800-218A",
+      "control_id": "PW.8.2-PS",
+      "control_name": "Test for security vulnerabilities — synthetic data adversarial testing",
+      "tier": "Foundational",
+      "scope": "Build",
+      "notes": "Conduct adversarial testing of synthetic data for membership inference, attribute inference, and reconstruction attacks to validate privacy claims"
+    },
+    {
+      "framework": "NIST SP 800-218A",
+      "control_id": "RV.3.1-PS",
+      "control_name": "Analyse root causes — synthetic data failure analysis",
+      "tier": "Foundational",
+      "scope": "Build",
+      "notes": "When model failures trace to synthetic training data, conduct root cause analysis of the generation process, source data, and privacy mechanism"
+    },
+    {
+      "framework": "FedRAMP",
+      "control_id": "SI-4",
+      "control_name": "System Monitoring — synthetic data quality",
+      "tier": "Foundational",
+      "scope": "Build",
+      "notes": "Monitor synthetic data outputs for quality, privacy preservation, and absence of sensitive pattern leakage"
+    },
+    {
+      "framework": "FedRAMP",
+      "control_id": "CA-7",
+      "control_name": "Continuous Monitoring — synthetic data drift",
+      "tier": "Foundational",
+      "scope": "Build",
+      "notes": "Include synthetic data quality metrics in continuous monitoring; track for privacy degradation and bias drift"
+    },
+    {
+      "framework": "FedRAMP",
+      "control_id": "RA-5",
+      "control_name": "Vulnerability Scanning — synthetic data risks",
+      "tier": "Foundational",
+      "scope": "Build",
+      "notes": "Include synthetic data re-identification and pattern leakage in vulnerability assessment"
+    },
+    {
+      "framework": "DORA",
+      "control_id": "Art. 9",
+      "control_name": "Protection and Prevention — synthetic data privacy",
+      "tier": "Foundational",
+      "scope": "Build",
+      "notes": "Implement privacy controls for synthetic data generation; validate privacy preservation and absence of sensitive pattern leakage from source financial data"
+    },
+    {
+      "framework": "DORA",
+      "control_id": "Art. 24–27",
+      "control_name": "Resilience Testing — synthetic data testing",
+      "tier": "Foundational",
+      "scope": "Build",
+      "notes": "Include synthetic data re-identification and privacy testing in resilience testing programme; test for linkage attacks and attribute inference"
+    },
+    {
+      "framework": "DORA",
+      "control_id": "Art. 13",
+      "control_name": "Learning and Evolving — synthetic data improvement",
+      "tier": "Foundational",
+      "scope": "Build",
+      "notes": "Apply lessons learned from synthetic data privacy failures; update generation processes and validation controls"
+    }
+  ],
+  "tools": [
+    {
+      "name": "Synthetic Data Vault",
+      "type": "open-source",
+      "url": "https://sdv.dev"
+    },
+    {
+      "name": "ML Privacy Meter",
+      "type": "open-source",
+      "url": "https://github.com/privacytrustlab/ml_privacy_meter"
+    },
+    {
+      "name": "Gretel AI",
+      "type": "commercial",
+      "url": "https://gretel.ai"
+    },
+    {
+      "name": "ARX Anonymisation Tool",
+      "type": "open-source",
+      "url": "https://arx.deidentifier.org"
+    },
+    {
+      "name": "ARX Data Anonymization Tool",
+      "type": "open-source",
+      "url": "https://arx.deidentifier.org"
+    },
+    {
+      "name": "TensorFlow Privacy",
+      "type": "open-source",
+      "url": "https://github.com/tensorflow/privacy"
+    },
+    {
+      "name": "OpenDP",
+      "type": "open-source",
+      "url": "https://github.com/opendp/opendp"
+    },
+    {
+      "name": "SDV (Synthetic Data Vault)",
+      "type": "open-source",
+      "url": "https://github.com/sdv-dev/SDV"
+    },
+    {
+      "name": "Gretel.ai",
+      "type": "commercial",
+      "url": "https://gretel.ai"
+    },
+    {
+      "name": "ARX",
+      "type": "open-source",
+      "url": "https://github.com/arx-deidentifier/arx"
+    },
+    {
+      "name": "SDV",
+      "type": "open-source",
+      "url": "https://github.com/sdv-dev/SDV"
+    },
+    {
+      "name": "Anonymeter",
+      "type": "open-source",
+      "url": "https://github.com/statice/anonymeter"
+    },
+    {
+      "name": "ARX Data Anonymization",
+      "type": "open-source",
+      "url": "https://arx.deidentifier.org"
+    },
+    {
+      "name": "Microsoft Presidio",
+      "type": "open-source",
+      "url": "https://github.com/microsoft/presidio"
+    }
+  ],
+  "incidents": [
+    {
+      "name": "Synthetic data re-identification — de-anonymized patients from synthetic health records",
+      "url": "https://github.com/emmanuelgjr/GenAI-Security-Crosswalk/blob/main/data/incidents.json",
+      "year": 2025,
+      "incident_id": "INC-040"
+    },
+    {
+      "name": "Stability AI synthetic CSAM generation — training data and output safety failures",
+      "url": "https://github.com/emmanuelgjr/GenAI-Security-Crosswalk/blob/main/data/incidents.json",
+      "year": 2024,
+      "incident_id": "INC-049"
+    }
+  ],
+  "crossrefs": {
+    "dsgai_2026": [
+      "DSGAI08",
+      "DSGAI18"
+    ],
+    "llm_top10": [
+      "LLM02",
+      "LLM01",
+      "LLM08",
+      "LLM09",
+      "LLM03"
+    ],
+    "agentic_top10": [
+      "ASI03",
+      "ASI06",
+      "ASI09"
+    ]
+  },
+  "changelog": [
+    {
+      "date": "2026-03-27",
+      "version": "1.0.0",
+      "change": "Initial entry — generated from GenAI Security Crosswalk v1.5.1 mapping files",
+      "author": "emmanuelgjr"
+    }
+  ]
+}