The AI Red Teaming Wiki
The definitive knowledge base for AI red teaming, prompt injection, and LLM security.
3000+ expert-level guides covering prompt injection techniques, jailbreak research, agent exploitation, MCP attack surfaces, RAG poisoning, model extraction, adversarial ML, and end-to-end red team engagements. Built by and for AI security researchers.
Explore Topics
36 parts, 1295 sections, 3000+ in-depth guides.
LLM Internals & Exploit Primitives
Transformer architecture, tokenizer exploitation, alignment bypass, embedding attacks.
Prompt Injection & Jailbreaks
Advanced injection techniques, automated jailbreak research, multimodal attack vectors.
Agent & Agentic Exploitation
AI agent attacks, multi-agent/A2A protocol exploitation, MCP tool surface attacks.
RAG, Data & Training Attacks
RAG pipeline poisoning, training data attacks, model extraction and IP theft.
Infrastructure & Supply Chain
Model serialization RCE, AI infra exploitation, application security patterns.
Recon & Tradecraft
LLM fingerprinting, system prompt extraction, AI-specific threat modeling.
Exploit Dev & Tooling
Custom adversarial tools, red team C2 frameworks, continuous automated red teaming.
Capstone: Full Engagement
End-to-end red team engagement methodology, reporting, and remediation.
Recently Updated
Agent Delegation Attacks
advancedExploiting multi-agent delegation patterns to achieve lateral movement, privilege escalation, and command-and-control through impersonation and insecure inter-agent communication.
Updated 2026-03-24
Agent Goal Hijacking
intermediateTechniques for redirecting AI agent objectives through poisoned inputs, indirect prompt injection, and multi-step manipulation -- the #1 ranked risk in OWASP's 2026 Agentic Top 10.
Updated 2026-03-24
Agent Identity and Credential Theft
advancedExploiting how AI agents authenticate to external services -- credential theft through agent manipulation, MFA bypass, and impersonation attacks including BodySnatcher and CVE-2025-64106.
Updated 2026-03-24
Agent Memory Poisoning
advancedTechniques for injecting malicious content into agent memory systems -- conversation history, RAG stores, and vector databases -- to achieve persistent cross-session compromise.
Updated 2026-03-24
Agent Supply Chain Attacks
intermediateCompromising AI agents through poisoned packages, backdoored MCP servers, malicious model registries, and weaponized agent frameworks -- including the Postmark MCP breach and NullBulge campaigns.
Updated 2026-03-24
Stay Updated
Get the latest AI red teaming research, techniques, and tool releases delivered to your inbox.