# Jailbreak Defense on GCG adversarial prompts (test)

Source: [Jailbreak Defense on GCG adversarial prompts (test)](<https://www.sota2.com/research/sota/jailbreak-defense-on-gcg-adversarial-prompts-test>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-02-26T05:40:31.919Z

Evaluation of model robustness and defense mechanisms against adversarial jailbreak prompts generated using the GCG method, measured by Defense Success Rate (DSR) and latency.

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 Defense |
| Dataset (including any stated split) | GCG adversarial prompts (test) |
| Metric | DSR |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 6 of 6 |
| 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

- [DSR](<https://www.sota2.com/research/sota/jailbreak-defense-on-gcg-adversarial-prompts-test.md?metric=DSR>): Higher is better; unit: PERCENT.

- [Avg Latency](<https://www.sota2.com/research/sota/jailbreak-defense-on-gcg-adversarial-prompts-test.md?metric=Avg%20Latency>): Lower is better; unit: TIME.

## Recorded results

| Method | DSR | Result date | Report paper |
| --- | --- | --- | --- |
| Backtranslation w/o early termination | 98 | Not recorded | 68fa738f1e209131cea5d3a7 |
| Backtranslation w/ early termination | 96 | Not recorded | 68fa738f1e209131cea5d3a7 |
| SmoothLLM | 92 | Not recorded | 68fa738f1e209131cea5d3a7 |
| Paraphrase | 84 | Not recorded | 68fa738f1e209131cea5d3a7 |
| Response Check | 30 | Not recorded | 68fa738f1e209131cea5d3a7 |
| No defense | 8 | Not recorded | 68fa738f1e209131cea5d3a7 |

### Backtranslation w/o early termination

DSR: 98. Result ID: 699300653a6c31484912f75d.

[Reporting paper](<https://www.sota2.com/research/paper/68fa738f1e209131cea5d3a7.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa738f1e209131cea5d3a7.md>)

[Implementation](<https://github.com/YihanWang617/LLM-Jailbreaking-Defense-Backtranslation>)

Recorded evaluation configuration: {"Target Model":"Vicuna-13B","early termination":false}. This record does not assert independent reproduction.

### Backtranslation w/ early termination

DSR: 96. Result ID: 699300653a6c31484912f75e.

[Reporting paper](<https://www.sota2.com/research/paper/68fa738f1e209131cea5d3a7.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa738f1e209131cea5d3a7.md>)

[Implementation](<https://github.com/YihanWang617/LLM-Jailbreaking-Defense-Backtranslation>)

Recorded evaluation configuration: {"Target Model":"Vicuna-13B","early termination":true}. This record does not assert independent reproduction.

### SmoothLLM

DSR: 92. Result ID: 699300653a6c31484912f75a.

[Reporting paper](<https://www.sota2.com/research/paper/68fa738f1e209131cea5d3a7.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73961e209131cea5d658.md>)

Recorded evaluation configuration: {"Target Model":"Vicuna-13B"}. This record does not assert independent reproduction.

### Paraphrase

DSR: 84. Result ID: 699300653a6c31484912f75b.

[Reporting paper](<https://www.sota2.com/research/paper/68fa738f1e209131cea5d3a7.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa707f1e209131cea4b7c6.md>)

Recorded evaluation configuration: {"Target Model":"Vicuna-13B"}. This record does not assert independent reproduction.

### Response Check

DSR: 30. Result ID: 699300653a6c31484912f75c.

[Reporting paper](<https://www.sota2.com/research/paper/68fa738f1e209131cea5d3a7.md>)

Recorded evaluation configuration: {"Target Model":"Vicuna-13B"}. This record does not assert independent reproduction.

### No defense

DSR: 8. Result ID: 699300653a6c31484912f759.

[Reporting paper](<https://www.sota2.com/research/paper/68fa738f1e209131cea5d3a7.md>)

Recorded evaluation configuration: {"Target Model":"Vicuna-13B"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.