attack-taxonomy

name	Attack Taxonomy
description	Complete taxonomy of 80+ attack plugins organized by category, technique, and OWASP mapping
version	1.0.0

Attack Taxonomy

Purpose

Provides a structured classification of all PromptArmor attack plugins. Used by agents to understand what attacks are available, how they work, and which vulnerabilities they target.

When to Use

• When planning which attacks to run
• When explaining findings to users
• When mapping attacks to OWASP or compliance frameworks
• When selecting mutation strategies

Attack Categories

Category	Count	Techniques
Jailbreak	13	Direct instruction bypass, persona switching, encoding, delimiter injection, context manipulation
Injection	5	Direct prompt injection, indirect injection via content, context compliance, system override
Tool Abuse	12	SQL injection, SSRF, file access, code execution, privilege escalation, data exfiltration
Harmful	18	Violence, crime, hate speech, self-harm, drugs, weapons, cybercrime, misinformation
Bias	5	Age, gender, race, disability, religion discrimination
PII	4	Direct disclosure, session leaks, social engineering, database leaks
Compliance	4	HIPAA, COPPA, FERPA, GDPR violations
Agentic	3	Memory poisoning, cross-session leaks, goal misalignment
RAG	3	Document exfiltration, source attribution, context poisoning
Other	13+	Hallucination, imitation, contracts, competitors, debug access, DoS

Reference Files

• Jailbreak Techniques
• Injection Patterns
• Tool Abuse Vectors
• Encoding Mutations