What is a DAN jailbreak?

DAN (Do Anything Now) variants bypass LLM safety by forcing the model into an unrestricted alter-ego. The database includes 15 DAN variants.

How are jailbreaks categorized?

Categories include DAN variants, character roleplay, encoding tricks, context manipulation, multi-turn attacks, system prompt extraction, output format exploits, language switching, fictional scenarios, token manipulation, delimiter injection, and other.

What mitigation strategies are listed?

Each entry includes specific mitigations such as input filtering, output validation, prompt monitoring, and model alignment updates.

🔐 LLM Jailbreak Database 100 known attack patterns

Interactive repository of jailbreak techniques — filter, search, and understand mitigations.

📊 Showing 100 / 100 patterns 🔍 all categories

Name	Category	Description	Example (redacted)	Models	Severity	Detection	Mitigation