Know how the guardrails fail.
A practitioner reference for LLM jailbreak techniques. Working bypasses, model behaviors they exploit, the patches that did and didn't fix them — written for AI red teamers who need to know what's still landing today.
Enter the archive →Latest entries
How to Detect LLM Jailbreak Attempts: A 5-Layer Stack
DefenseOpen Source LLM Security Tools: Licensing and Cost
ToolingJailbreakBench Dataset Explained: JBB-Behaviors vs AdvBench
ToolingPayload Splitting: How Fragmented Prompts Evade LLM Filters
TechniquesLLM Red Teaming Tools Compared: Garak, PyRIT & More
ToolingLethal Trifecta and CaMeL: Containing Agent Injection
DefenseBest LLM Guardrail Tools 2026: A Practitioner's Comparison
DefenseHow LLM Jailbreaks Work: Techniques and Success Rates
FoundationsDAN, Evil Confidant & Skeleton Key Jailbreaks Explained
TechniquesLLM Defense Stack: Guardrails, Tool Scoping, and Egress
DefenseStart here
23 writeups is a lot to land in cold. These are the entry points most people want, in the order they usually want them.
- Why jailbreaks work at all
Competing objectives and mismatched generalization — the two failure modes every technique on this site exploits.
- DAN, Evil Confidant & Skeleton Key
The persona jailbreaks every red team meets first: how each one worked, why it worked, and what replaced them.
- LLM red teaming tools compared
Garak, PyRIT, Promptfoo and Giskard compared by licence, CI fit, and what none of them cover.
- Jailbreak benchmarks & datasets
JailbreakBench, HarmBench, AdvBench and StrongREJECT: which number you are reading, and which one to produce.
- Payload splitting & encoding
Splitting a blocked request across turns or encodings so no single message trips the input filter.
- PAIR vs GCG vs TAP
Which automated jailbreak framework to run: threat model fit, query budget, transferability, and compute cost.
Independent, specialist, and free to read
Jailbreaks FYI publishes focused, sourced guides on a single topic. No paywall, no account, no ad tracking.
Jailbreaks FYI — in your inbox
Working LLM jailbreak techniques, sourced and dated — delivered when there's something worth your inbox.
No spam. Unsubscribe anytime.