Awesome-Jailbreak-on-LLMs
A curated academic collection of jailbreak attack and defense research on large and vision-language models, with papers, code, datasets and evaluations.
Awesome-Jailbreak-on-LLMs is a research-oriented curated list covering jailbreak methods on large language models and vision-language models, along with the papers, code, datasets and evaluations behind them. Entries are organized under jailbreak attacks, split into categories such as attacks on reasoning models, black-box and white-box attacks, multi-turn attacks, attacks on retrieval-augmented systems and multi-modal attacks, and under jailbreak defenses, split into learning-based defenses, strategy-based defenses, guard models and moderation APIs, plus sections on evaluation and analysis and on applications.
Each paper entry links to its source and, where available, its released code, and the list is actively maintained with contributions welcomed through issues and pull requests. It is licensed under MIT and intended as a reference for AI-safety researchers studying how models can be attacked and defended.