OWASP LLM Top 10
Authoritative list of critical risks for LLM applications—start here for scoping and testing.
LLM Prompt Injection & Jailbreaks
Resources for red teamers and AppSec practitioners testing LLM applications
⚠️ Authorized Testing Only: Use these techniques only on systems you own or have explicit permission to test.
Prompt injection testing is valuable for securing LLM-powered apps—scope your engagements and document findings responsibly.
Why Red Teamers Care
LLM jailbreaks and prompt injection are attack vectors that bypass safety guardrails, exfiltrate data, or elicit unintended behavior from AI systems. As LLMs are integrated into apps and APIs, these techniques belong in your testing toolkit—alongside OWASP guidance and proper scoping.
Use the resources below to understand current bypass patterns, plan LLM security assessments, and harden applications against adversarial input.
Jailbreak Techniques & Resources
Jailbreak Chat (Community)
Community-driven jailbreak prompts and techniques; useful for understanding current bypass patterns.
Prompt Injection OWASP Cheat Sheet
Practical prompt injection attack vectors and defensive guidance.
Prompt Injection Everywhere
Comprehensive guide to indirect prompt injection, data exfiltration, and adversarial prompting.
Universal LLM Jailbreak Patterns
Research repo: cross-model techniques for GPT-4, Claude, and other LLMs.
Simon Willison's Prompt Injection Notes
Practical notes on prompt injection from a security-focused developer.
Research & Community Resources
OWASP LLM Top 10
Critical risks for LLM applications—test scope
MITRE ATLAS (Adversarial ML)
Adversarial threats to machine learning systems
Simon Willison: Prompt Injection
Practical security research on injection
arXiv: LLM Security
Academic research on AI safety
⚠️ Important Notice
These resources are provided for educational and research purposes only. Unauthorized use of jailbreak techniques may violate terms of service and applicable laws. Always obtain proper authorization before conducting security testing on AI systems you do not own or have explicit permission to test.