Harmful Prompts
Benchmarks
| Task Name | Dataset Name | SOTA Result | Trend | |
|---|---|---|---|---|
| Jailbreak evaluation | Harmful Prompts Curated April 13, 2023 | Bad Bot Rate0 | 61 | |
| Safety Evaluation | Harmful Prompts | Harmful Score8.3 | 40 | |
| Safety Evaluation | Harmful Prompts | ASR (Raw)2 | 15 | |
| Stealthiness Evaluation | Harmful prompts (evaluated on 3 LLMs and 4 guard LLMs) | Mean Perplexity3.23 | 10 | |
| Safety Evaluation | Harmful Prompts Text-only baseline | Text ASR0 | 10 | |
| Jailbreak Attack | 100 Harmful Prompts | ASR (K=2)55 | 9 | |
| Jailbreak Attack | Harmful Prompts model-averaged | Model Averaged ASR (0.8)4.8 | 8 | |
| Adversarial Attack | Harmful Prompts | ASR70.3 | 8 | |
| Jailbreak Attack | Harmful Prompts 8-model (test) | ASR92.4 | 5 | |
| Jailbreak Attack | Harmful Prompts Victim: GPT-4o | Runtime (H:MM)4 | 5 | |
| Jailbreak Attack | Harmful Prompts SDXL | Attack Success Rate (ASR)88 | 4 |
Showing 11 of 11 rows