Skip to content

Latest commit

 

History

History
125 lines (95 loc) · 7.29 KB

File metadata and controls

125 lines (95 loc) · 7.29 KB

OWASP GenAI Crosswalk — Master Cross-Reference

This is the single navigation index across all three OWASP GenAI source lists. Read left to right: an LLM vulnerability → its agentic amplification → its data security dimension.

How to use: Find the risk you care about in any column, then follow the row to see how it surfaces in the other lists and which framework files cover it.


Cross-reference table

LLM Top 10 2026 Agentic Top 10 2026 DSGAI 2026 Primary frameworks
LLM01 Prompt Injection ASI01 Agent Goal Hijack DSGAI01 Sensitive Data Leakage MITRE ATLAS · AIUC-1 B001/B005 · NIST AI RMF · ASVS
LLM01 Prompt Injection ASI02 Tool Misuse DSGAI12 Unsafe NL Data Gateways MITRE ATLAS · AIUC-1 B006 · ISA 62443 · STRIDE
LLM02 Sensitive Information Disclosure ASI03 Identity & Privilege Abuse DSGAI01 Sensitive Data Leakage ISO 27001 · AIUC-1 A · SOC 2 · PCIDSS
LLM02 Sensitive Information Disclosure ASI03 Identity & Privilege Abuse DSGAI02 Agent Identity & Credential Exposure OWASP NHI · AIUC-1 B007 · NIST CSF 2.0 · ISO 27001
LLM03 Excessive Agency ASI01 Agent Goal Hijack DSGAI06 Tool Plugin & Agent Data Exchange AIUC-1 B006 · ISA 62443 · NIST AI RMF
LLM03 Excessive Agency ASI10 Rogue Agents DSGAI16 Endpoint & Browser Overreach AIUC-1 C · MITRE ATLAS · STRIDE · EU AI Act
LLM04 Supply Chain ASI04 Agentic Supply Chain DSGAI04 Data Model & Artifact Poisoning NIST SP 800-218A · SAMM
LLM05 Data and Model Poisoning ASI06 Memory & Context Poisoning DSGAI04 Data Model & Artifact Poisoning MITRE ATLAS · AIUC-1 A · CIS Controls · AITG
LLM05 Data and Model Poisoning ASI06 Memory & Context Poisoning DSGAI21 Disinformation via Data Poisoning STRIDE · ENISA · EU AI Act
LLM06 Unbounded Consumption ASI08 Cascading Failures DSGAI17 Data Availability & Resilience Failures AIUC-1 D · ISA 62443 · NIST SP 800-82 · CIS Controls
LLM07 Misinformation ASI09 Human-Agent Trust Exploitation DSGAI21 Disinformation via Data Poisoning AIUC-1 C/F · EU AI Act · ENISA · NIST AI RMF
LLM08 Hidden Context Exposure ASI01 Agent Goal Hijack DSGAI15 Over-Broad Context Windows AIUC-1 B003/B009 · ISO 27001 · CIS Controls · ASVS
LLM09 Vector and Embedding Weaknesses ASI06 Memory & Context Poisoning DSGAI13 Vector Store Platform Security MITRE ATLAS · AIUC-1 A · NIST AI RMF · CWE/CVE
LLM10 Improper Output Handling ASI02 Tool Misuse DSGAI05 Data Integrity & Validation Failures ASVS · CIS Controls · NIST CSF 2.0 · AIUC-1 B009
ASI05 Unexpected Code Execution DSGAI12 Unsafe NL Data Gateways AIUC-1 B006 · ASVS · CWE/CVE
ASI07 Insecure Inter-Agent Comms DSGAI02 Agent Identity & Credential Exposure OWASP NHI · AIUC-1 E · ISA 62443 · ISO 27001
DSGAI03 Shadow AI & Unsanctioned Data Flows ISO 27001 · ENISA · CIS Controls
DSGAI07 Data Governance & Lifecycle ISO 42001 · EU AI Act · NIST AI RMF
DSGAI08 Non-Compliance & Regulatory Violations EU AI Act · ISO 42001 · PCIDSS · SOC 2
DSGAI09 Multimodal Cross-Channel Leakage ISO 27001 · AIUC-1 A · CIS Controls · ENISA
DSGAI10 Synthetic Data & Anonymization Pitfalls NIST AI RMF · EU AI Act · ENISA
DSGAI11 Cross-Context Conversation Bleed ASVS · ISO 27001 · NIST CSF 2.0 · AIUC-1 A
DSGAI14 Excessive Telemetry & Monitoring Leakage ISO 27001 · CIS Controls · SOC 2 · NIST CSF 2.0
DSGAI18 Inference & Data Reconstruction MITRE ATLAS · ENISA · AIUC-1 A
DSGAI19 Human-in-Loop & Labeler Overexposure EU AI Act · NIST AI RMF
DSGAI20 Model Exfiltration & IP Replication MITRE ATLAS · CWE/CVE · ISO 27001 · AIUC-1 B

Agentic Skills Top 10 (AST) — the skill/behaviour layer

A fourth OWASP source list, registered here as AST01AST10. It covers the skill layer: the SKILL.md and manifest files telling an agent how to orchestrate tools, as distinct from MCP, which defines what tools exist.

Attribution. The Agentic Skills Top 10 is a sibling OWASP project with its own leads and contributors. It is credited to that project exactly as the LLM Top 10 and the Agentic Top 10 are, and appears here as an external source list. Canonical spec: https://owasp.org/www-project-agentic-skills-top-10/

Names, severities and MAESTRO layers are transcribed from that page (verified 2026-08-28). Framework mappings are not yet authored — that is expert work, tracked as T-A10-03. These rows are an index, not a mapping.

ID Risk Severity MAESTRO layers
AST01 Malicious Skills Critical 7, 3, 6, 4, 5
AST02 Supply Chain Compromise Critical 7, 3, 6, 4
AST03 Over-Privileged Skills High 6, 4, 3, 7
AST04 Insecure Metadata High 7, 3, 4, 6
AST05 Untrusted External Instructions High 3, 2, 7, 6
AST06 Weak Isolation High 4, 6, 3
AST07 Update Drift Medium 4, 6, 7
AST08 Poor Scanning Medium 5, 6, 3
AST09 No Governance Medium 6, 7, 5
AST10 Cross-Platform Reuse Medium 7, 3, 6

Referenced but not yet mapped

These frameworks are relevant to the risks above and are not part of this crosswalk. They were previously listed in the navigation column, which implied a mapping file that does not exist. They are recorded here instead so the omission is visible rather than silent:

Framework Why it is relevant Status
MITRE ATT&CK Enterprise adversary techniques that AI attacks chain into Not mapped
ISO/IEC 27701 Privacy information management, extending ISO 27001 Not mapped
CycloneDX ML-BOM Machine-readable model and dataset inventory Not mapped
BSIMM Software security maturity measurement Not mapped
COBIT IT governance and management objectives Not mapped

Adding any of them is a maintainer decision about scope, not an oversight — see roadmap issue #19.


OT / ICS lens

For OT environments, these entries carry the highest blast radius:

Risk Why it matters in OT Primary OT frameworks
LLM01 / ASI01 NL-to-SCADA interfaces make prompt injection a process disruption risk ISA/IEC 62443 · NIST SP 800-82
LLM06 / ASI08 Cascading failures in OT can affect physical safety systems ISA/IEC 62443 SR 7.1 · NIST SP 800-82
ASI02 Tool Misuse AI agents with OT tool access can trigger physical actions ISA/IEC 62443 · NIST SP 800-82
DSGAI12 LLM-to-SQL/historian gateways in OT expose process data ISA/IEC 62443 · NIST SP 800-82
DSGAI17 RAG failures in OT AI affect real-time operational decisions ISA/IEC 62443 SR 7.6 · NIST SP 800-82

Framework index

Jump directly to any framework mapping folder:

LLM Top 10 mappings → /llm-top10/

Agentic Top 10 mappings → /agentic-top10/

DSGAI 2026 mappings → /dsgai-2026/

Shared resources → /shared/

Machine-readable data → /data/


Last updated: 2026-Q1 | Maintained by the OWASP GenAI Data Security Initiative