Topics
Browse everything JailbreakDB publishes by category and tag: every topic covered here, with the most recent guides listed under each one.
Tags
- #llm-security 13
- #jailbreak 8
- #red-teaming 7
- #prompt-injection 6
- #taxonomy 3
- #guardrails 2
- #jailbreak-detection 2
- #adaptive-attacks 1
- #adversarial-attacks 1
- #adversarial-testing 1
- #agent-security 1
- #arms-race 1
- #base64 1
- #behavioral-monitoring 1
- #benchmark 1
- #classifier 1
- #classifier-evasion 1
- #content-filter 1
- #content-filter-bypass 1
- #crescendo 1
- #cvd 1
- #dan 1
- #defense 1
- #detection-evasion 1
- #encoding-attacks 1
- #evaluation 1
- #gcg 1
- #gradient-attacks 1
- #harmbench 1
- #in-context-learning 1
- #input-filtering 1
- #jailbreak-history 1
- #jailbreakbench 1
- #long-context 1
- #many-shot-jailbreaking 1
- #multi-turn 1
- #obfuscation 1
- #persona 1
- #research 1
- #research-ethics 1
- #responsible-disclosure 1
- #rlhf 1
- #roleplay-jailbreak 1
- #security-testing 1
- #strongreject 1
- #transferability 1
- #unicode 1
- #universal-suffix 1
- #vendor-evaluation 1
- #vulnerability-disclosure 1
Categories
technique 6 posts
- Prompt Injection vs Jailbreak: Two Different Attacks on LLMsPrompt injection and jailbreak get conflated constantly, but they hit different layers and need different defenses. Here is how to tell them apart.
- LLM Jailbreak Techniques Explained: 7 Attack FamiliesA technical breakdown of LLM jailbreak techniques by attack category, from role-play bypasses to multilingual exploits and multi-turn escalation.
- Crescendo Multi-Turn Jailbreaks: Why Defenses Miss ThemCrescendo multi-turn jailbreaks evade per-message defenses by exploiting conversation context, requiring session-level classification and monitoring.
- Encoding and Obfuscation Jailbreaks: The Filter-Model GapEncoding and obfuscation jailbreaks exploit representation gaps between content filters and LLMs; normalization and decoding can help detect them.
- Roleplay and Persona Jailbreaks: Why They Often FailRoleplay and persona jailbreaks such as DAN use fictional framing, but modern safety training blocks many variants on frontier models.
- Universal Adversarial Suffixes: How the GCG Attack WorksGreedy Coordinate Gradient finds nonsense token suffixes that jailbreak aligned LLMs and transfer to closed models, plus the defenses that blunt them.
research 3 posts
- How Jailbreak Benchmarks Measure Success: JailbreakBenchHow to read a jailbreak attack-success-rate: the behaviour set, attacker harness, judge and target that decide the number, and why two papers disagree.
- Jailbreak Detection-Evasion Arms Race: How Attackers AdaptJailbreak detection and evasion form an arms race in which attackers probe classifier boundaries and defenders use layered safeguards.
- LLM Jailbreak Taxonomy 2026: How the Techniques ClusterSix years of jailbreak research has produced a messy literature. This taxonomy organizes working techniques by the behavioral property each one exploits.
defense 2 posts
- AI Jailbreak Detection Tools: Six Options Worth ShortlistingA technical buyer's guide to jailbreak detectors, prompt-injection filters, cloud guardrails, and the coverage gaps vendors tend to bury in documentation.
- How to Detect Jailbreak Prompts: The Three LayersThe three jailbreak detection layers — input classifiers, output monitoring, and in-model sentinels — what each one catches, and where each is evaded.
tools 2 posts
- LLM Security Testing Platform: Buyer's Guide for 2026Picking an LLM security testing platform: open-source harness vs commercial suite, who got acquired, and five checks that expose weak tools.
- Best LLM Red Team Tools: Garak, PyRIT, and PromptfooThe best LLM red team tools are compared across attack coverage, multi-turn depth, CI integration, strengths, limitations, and ideal use cases.