A system prompt to prevent prompt leakage and adversarial attacks in GPTs.
Dataset for classifying prompts as jailbreak or benign to enhance LLM safety.
A collection of state-of-the-art jailbreak methods for LLMs, including papers, codes, datasets, and analyses.
A dataset of 15,140 ChatGPT prompts, including 1,405 jailbreak prompts, collected from various platforms for research purposes.
