OWASP LLM Top 10 Compliance Tracker

The OWASP Top 10 for LLM Applications identifies the most critical security risks for applications using Large Language Models. Each risk category below is mapped to real CVEs affecting AI/ML packages in production.

3792
CVEs Mapped
10
Controls with CVEs
4746
Total Mappings

Controls & Mapped Vulnerabilities

LLM01

Prompt Injection

133 CVEs
CRITICAL
CVE-2026-61539CVSS 10.0

Xinference: eval() on LLM output enables RCE

CRITICAL
CVE-2026-54769CVSS 10.0

Langroid: prompt injection to RCE via broken eval() sandbox

CRITICAL
CVE-2026-61447CVSS 10.0

PraisonAI: RCE via unsandboxed LLM code execution

+ 130 more CVEs mapped to this control

LLM02

Sensitive Information Disclosure

346 CVEs
CRITICAL
CVE-2026-61447CVSS 10.0

PraisonAI: RCE via unsandboxed LLM code execution

CRITICAL
CVE-2026-34938CVSS 10.0

praisonaiagents: sandbox bypass enables full host RCE

CRITICAL
CVE-2026-33663CVSS 10.0

n8n: member role steals plaintext HTTP credentials

+ 343 more CVEs mapped to this control

LLM03

Supply Chain Vulnerabilities

560 CVEs
CRITICAL
CVE-2026-111057CVSS 10.0

Mercury: regex blocklist bypass allows unauthenticated RCE

CRITICAL
CVE-2026-94603CVSS 10.0

Podman: checkpoint images override sandbox on run

CRITICAL
CVE-2024-39700CVSS 10.0

JupyterLab extension-template: CI workflow RCE

+ 557 more CVEs mapped to this control

LLM04

Data and Model Poisoning

200 CVEs
CRITICAL
CVE-2025-53002CVSS 9.8

LLaMA-Factory: RCE via unsafe checkpoint deserialization

CRITICAL
CVE-2026-55450CVSS 9.3

Langflow: unauthenticated upload → DoS + path disclosure

CRITICAL
CVE-2026-27493CVSS 9.0

n8n: Code Injection enables RCE

+ 197 more CVEs mapped to this control

LLM05

Improper Output Handling

316 CVEs
CRITICAL
CVE-2026-35307CVSS 10.0

Oracle Coherence: unauthenticated RCE, CVSS 10.0

CRITICAL
CVE-2026-61539CVSS 10.0

Xinference: eval() on LLM output enables RCE

CRITICAL
CVE-2025-15379CVSS 10.0

MLflow: RCE via unsanitized model dependency specs

+ 313 more CVEs mapped to this control

LLM06

Excessive Agency

836 CVEs
CRITICAL
CVE-2026-33663CVSS 10.0

n8n: member role steals plaintext HTTP credentials

CRITICAL
GHSA-jqmf-mx4f-hfr6CVSS 10.0

Vibe-Trading: unauth agent RCE via shell tools, exec

CRITICAL
CVE-2026-46695CVSS 10.0

Boxlite: read-only bypass enables host code execution

+ 833 more CVEs mapped to this control

LLM07

System Prompt Leakage

388 CVEs
CRITICAL
CVE-2026-34938CVSS 10.0

praisonaiagents: sandbox bypass enables full host RCE

CRITICAL
CVE-2025-53767CVSS 10.0

Azure OpenAI: SSRF EoP, no auth required (CVSS 10)

CRITICAL
CVE-2025-71338CVSS 10.0

Flowise: unauthenticated file write enables RCE

+ 385 more CVEs mapped to this control

LLM08

Vector and Embedding Weaknesses

366 CVEs
CRITICAL
CVE-2025-14931CVSS 10.0

smolagents: RCE via pickle deserialization in executor

CRITICAL
CVE-2026-34938CVSS 10.0

praisonaiagents: sandbox bypass enables full host RCE

CRITICAL
CVE-2026-54769CVSS 10.0

Langroid: prompt injection to RCE via broken eval() sandbox

+ 363 more CVEs mapped to this control

LLM09

Misinformation

20 CVEs
CRITICAL
CVE-2025-61260CVSS 9.8

OpenAI Codex CLI: RCE via malicious MCP config files

CRITICAL
CVE-2024-39236CVSS 9.8

Gradio: code injection via component metadata (CVSS 9.8)

HIGH
GHSA-m3mh-3mpg-37hwCVSS 8.6

OpenClaw: .npmrc hijack enables RCE on plugin install

+ 17 more CVEs mapped to this control

LLM10

Unbounded Consumption

126 CVEs
CRITICAL
CVE-2026-57116CVSS 9.8

PraisonAI: unauthenticated AgentOS chat API drives agents

CRITICAL
CVE-2026-57118CVSS 9.8

PraisonAI agents: unauthenticated access to agent endpoints

HIGH
CVE-2026-92954CVSS 8.6

vm2: unhandled host Promise rejection crashes Node host

+ 123 more CVEs mapped to this control

Download Full Evidence Pack

Get the complete OWASP LLM Top 10 evidence pack with all CVE-to-control mappings, rationale, and audit-ready documentation. Exportable as CSV.

Get Evidence Pack