# Jailbreak Attack Robustness on GCG Source: [Jailbreak Attack Robustness on GCG]() Collection index: [llms.txt]() Data updated: 2026-05-07T08:08:35.391Z Evaluation of model robustness against adversarial jailbreak prompts using the GCG dataset, specifically measuring the Harmfulness Rate observed in the model's reasoning steps versus its final output. A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. | Scope | Recorded value | | --- | --- | | Task | Jailbreak Attack Robustness | | Dataset (including any stated split) | GCG | | Metric | Harmfulness Rate | | Unit | PERCENT | | Direction | Lower is better | | Results loaded | 21 of 21 | | Coverage | All results currently returned for this benchmark | Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol. ## Metrics - [Harmfulness Rate](): Lower is better; unit: PERCENT. - [Reasoning Harmfulness Rate](): Lower is better; unit: PERCENT. - [Final Answer Harmfulness Rate](): Lower is better; unit: PERCENT. ## Recorded results | Method | Harmfulness Rate | Result date | Report paper | | --- | --- | --- | --- | | RealSafe-R1 | 0 | Not recorded | 68fa7a721e209131cea8346b | | SafeKey | 0 | Not recorded | 68fa7a721e209131cea8346b | | Self-Reminder | 0 | Not recorded | 68fa7a721e209131cea8346b | | SAFEPATH-ZS | 0 | Not recorded | 68fa7a721e209131cea8346b | | ReasoningGuard | 0 | Not recorded | 68fa7a721e209131cea8346b | | RealSafe-R1 | 0 | Not recorded | 68fa7a721e209131cea8346b | | Self-Reminder | 0 | Not recorded | 68fa7a721e209131cea8346b | | SAFEPATH-ZS | 0 | Not recorded | 68fa7a721e209131cea8346b | | ReasoningGuard | 0 | Not recorded | 68fa7a721e209131cea8346b | | ThinkingI | 4 | Not recorded | 68fa7a721e209131cea8346b | | ThinkingI | 6 | Not recorded | 68fa7a721e209131cea8346b | | Paraphrase | 8 | Not recorded | 68fa7a721e209131cea8346b | | Paraphrase | 12 | Not recorded | 68fa7a721e209131cea8346b | | SmoothLLM | 14 | Not recorded | 68fa7a721e209131cea8346b | | No Defense | 16 | Not recorded | 68fa7a721e209131cea8346b | | SmoothLLM | 16 | Not recorded | 68fa7a721e209131cea8346b | | No Defense | 22 | Not recorded | 68fa7a721e209131cea8346b | ### RealSafe-R1 Harmfulness Rate: 0. Result ID: 69fc487947f34ce924dedbf8. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"R1-Qwen-14B"}. This record does not assert independent reproduction. ### SafeKey Harmfulness Rate: 0. Result ID: 69fc487947f34ce924dedbf9. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"R1-Qwen-14B"}. This record does not assert independent reproduction. ### Self-Reminder Harmfulness Rate: 0. Result ID: 69fc487947f34ce924dedbfb. [Reporting paper]() Recorded evaluation configuration: {"Model":"R1-Qwen-14B"}. This record does not assert independent reproduction. ### SAFEPATH-ZS Harmfulness Rate: 0. Result ID: 69fc487947f34ce924dedbfe. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"R1-Qwen-14B"}. This record does not assert independent reproduction. ### ReasoningGuard Harmfulness Rate: 0. Result ID: 69fc487947f34ce924dedbff. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"R1-Qwen-14B"}. This record does not assert independent reproduction. ### RealSafe-R1 Harmfulness Rate: 0. Result ID: 69fc487947f34ce924dedc01. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"R1-Qwen-32B"}. This record does not assert independent reproduction. ### Self-Reminder Harmfulness Rate: 0. Result ID: 69fc487947f34ce924dedc03. [Reporting paper]() Recorded evaluation configuration: {"Model":"R1-Qwen-32B"}. This record does not assert independent reproduction. ### SAFEPATH-ZS Harmfulness Rate: 0. Result ID: 69fc487947f34ce924dedc06. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"R1-Qwen-32B"}. This record does not assert independent reproduction. ### ReasoningGuard Harmfulness Rate: 0. Result ID: 69fc487947f34ce924dedc07. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"R1-Qwen-32B"}. This record does not assert independent reproduction. ### ThinkingI Harmfulness Rate: 4. Result ID: 69fc487947f34ce924dedbfd. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"R1-Qwen-14B"}. This record does not assert independent reproduction. ### ThinkingI Harmfulness Rate: 6. Result ID: 69fc487947f34ce924dedc05. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"R1-Qwen-32B"}. This record does not assert independent reproduction. ### Paraphrase Harmfulness Rate: 8. Result ID: 69fc487947f34ce924dedbfa. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"R1-Qwen-14B"}. This record does not assert independent reproduction. ### Paraphrase Harmfulness Rate: 12. Result ID: 69fc487947f34ce924dedc02. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"R1-Qwen-32B"}. This record does not assert independent reproduction. ### SmoothLLM Harmfulness Rate: 14. Result ID: 69fc487947f34ce924dedc04. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"R1-Qwen-32B"}. This record does not assert independent reproduction. ### No Defense Harmfulness Rate: 16. Result ID: 69fc487947f34ce924dedbf7. [Reporting paper]() Recorded evaluation configuration: {"Model":"R1-Qwen-14B"}. This record does not assert independent reproduction. ### SmoothLLM Harmfulness Rate: 16. Result ID: 69fc487947f34ce924dedbfc. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"R1-Qwen-14B"}. This record does not assert independent reproduction. ### No Defense Harmfulness Rate: 22. Result ID: 69fc487947f34ce924dedc00. [Reporting paper]() Recorded evaluation configuration: {"Model":"R1-Qwen-32B"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.