Jailbreak LLMs
StaleDescription
A dataset of 15,140 ChatGPT prompts including 1,405 jailbreak prompts from Reddit, Discord, and other platforms, providing a large-scale benchmark for LLM safety research and jailbreak detection.
Related Projects
Open-Prompt-Injection
439 · Python
An open-source benchmark for prompt injection attacks and defenses in LLMs, systematically evaluating the effectiveness of different attack strategies and defense mechanisms.
prompt-injectionbenchmarkllm-safety +2
JailTrickBench
162 · Python
Bag of Tricks for benchmarking jailbreak attacks on LLMs. NeurIPS 2024 paper providing empirical tricks for LLM jailbreaking with standardized evaluation.
benchmarkjailbreakllm-safety +2
Vigil
478 · Python
Vigil is an LLM security detection tool that identifies prompt injections, jailbreaks, and other potentially risky LLM inputs through multi-dimensional analysis for real-time safety protection.
prompt-injectionsecurityllm-safety +2
AgentShield Benchmark
21 · TypeScript
Open benchmark for AI agent security tools, evaluating prompt injection, data exfiltration, tool abuse, and provenance tracking.
securitybenchmarkai-safety +2