Vulnerability Database

70 AI/LLM vulnerabilities mapped to the OWASP LLM Top 10 (2025). Search, filter, and explore attack vectors, mitigations, and references.

Vulnerability Database

70 AI/LLM vulnerabilities mapped to the OWASP LLM Top 10 (2025). Search, filter, and explore attack vectors, mitigations, and references.

Vulnerability Database

Direct Prompt Injection

Indirect Prompt Injection

Training Data Extraction via Prompting

Markdown Image Exfiltration

Tool-Based Data Exfiltration

Malicious Model Supply Chain Attack

Compromised Model Hosting Platform

Training Data Poisoning

Backdoor Injection via Fine-Tuning

Cross-Site Scripting via LLM Output

SQL Injection via LLM-Generated Queries

Command Injection via LLM Code Generation

Unauthorized Tool Execution

Confused Deputy Attack on LLM Agents

Privilege Escalation via Tool Chaining

Cross-Tenant Data Leakage in Shared Vector DBs

Image-Based Prompt Injection

MCP Server Impersonation

MCP Tool Shadowing

MCP Prompt Injection via Tool Responses

Wallet Draining Attack

Adversarial Suffix Attack

Model Serialization Exploit

System Prompt Secrets Exposure

API Key Leakage via Model Output

MCP Cross-Server Data Exfiltration

MCP Rug Pull via Dynamic Tool Modification

Role-Play Jailbreak

Few-Shot Prompt Manipulation

Encoding-Based Injection

Link-Based Data Exfiltration

System Prompt Extraction

PII Leakage via Fine-Tuned Models

Dependency Confusion in ML Pipelines

Malicious Fine-Tuning Dataset

RAG Document Poisoning

Server-Side Request Forgery via LLM Output

Unsafe Deserialization of LLM Output

CSRF-Like Agent Manipulation

Autonomous Goal Drift

System Prompt Leakage via Instruction Following

Embedding Inversion Attack

Embedding Poisoning for RAG Manipulation

Confident Hallucination of Facts

Automated Disinformation Generation

Resource Exhaustion via Complex Prompts

Infinite Loop via Recursive Agent Tasks

Audio Adversarial Attack

MCP Credential Theft via Tool Arguments

Confidential Data in LLM Logs

Plugin/Extension Supply Chain Attack

Adversarial Perturbation of Training Images

Template Injection via LLM Output

Excessive File System Access via Agent

Multi-Modal Steganographic Injection

Batch Processing Resource Abuse

Synthetic Media Attribution Failure

Multi-Language Injection

Payload Splitting Injection

RLHF Reward Hacking

System Prompt Inference via Behavioral Analysis

Embedding Space Collision Attack

Deepfake Text Impersonation

Model Denial of Service via Adversarial Inputs

RAG Context Window Overflow

Context Window Manipulation

Knowledge Cutoff Exploitation

Sycophantic Bias Exploitation

Semantic Search Poisoning

Token-Level Denial of Service

Vulnerability Database

Direct Prompt Injection

Indirect Prompt Injection

Training Data Extraction via Prompting

Markdown Image Exfiltration

Tool-Based Data Exfiltration

Malicious Model Supply Chain Attack

Compromised Model Hosting Platform

Training Data Poisoning