Awesome-Jailbreak-on-LLMs

Why we included this project

People working on LLM safety, red-teaming, or alignment will find this a fast way to get oriented in how models get broken. It is a curated index of academic papers with links to code and datasets, sorted by attack type: black-box, white-box, multi-turn, multi-modal, and attacks on reasoning models, plus a separate section on defenses such as guard models and moderation APIs. The organization is the main draw: instead of chasing scattered arXiv listings, you get a categorized map of the field that is actively maintained and open to contributions. Teams evaluating guardrails or building evaluation suites can use it to track known attack vectors and the defenses proposed against them. It is a research resource, not a deployable tool, so treat it as a reference for methods and benchmarks, and for citations to follow up on.