# Jailbreak Attack on Advbench Harmful Behaviors (subset 100 samples)

Source: [Jailbreak Attack on Advbench Harmful Behaviors (subset 100 samples)](<https://www.sota2.com/research/sota/jailbreak-attack-on-advbench-harmful-behaviors-subset-100-samples>)

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

Data updated: 2026-02-26T05:35:06.888Z

Evaluation of model robustness against jailbreak attacks using a specific 100-sample subset of the Advbench Harmful Behaviors dataset.

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 |
| Dataset (including any stated split) | Advbench Harmful Behaviors (subset 100 samples) |
| Metric | Attack Success Rate AK |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 9 of 9 |
| 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

- [Attack Success Rate AK](<https://www.sota2.com/research/sota/jailbreak-attack-on-advbench-harmful-behaviors-subset-100-samples.md?metric=Attack%20Success%20Rate%20AK>): Higher is better; unit: PERCENT.

- [Attack Success Rate AG](<https://www.sota2.com/research/sota/jailbreak-attack-on-advbench-harmful-behaviors-subset-100-samples.md?metric=Attack%20Success%20Rate%20AG>): Higher is better; unit: PERCENT.

- [Attack Success Rate AA](<https://www.sota2.com/research/sota/jailbreak-attack-on-advbench-harmful-behaviors-subset-100-samples.md?metric=Attack%20Success%20Rate%20AA>): Higher is better; unit: PERCENT.

- [Attack Success Rate AU](<https://www.sota2.com/research/sota/jailbreak-attack-on-advbench-harmful-behaviors-subset-100-samples.md?metric=Attack%20Success%20Rate%20AU>): Higher is better; unit: PERCENT.

- [Attack Success Rate AP](<https://www.sota2.com/research/sota/jailbreak-attack-on-advbench-harmful-behaviors-subset-100-samples.md?metric=Attack%20Success%20Rate%20AP>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Attack Success Rate AK | Result date | Report paper |
| --- | --- | --- | --- |
| SCAV | 100 | Not recorded | 68fa7a761e209131cea83657 |
| SCAV | 99.85 | Not recorded | 68fa7a761e209131cea83657 |
| SCAV | 99.24 | Not recorded | 68fa7a761e209131cea83657 |
| CAVGAN | 98.98 | Not recorded | 68fa7a761e209131cea83657 |
| CAVGAN | 98.78 | Not recorded | 68fa7a761e209131cea83657 |
| CAVGAN | 95.51 | Not recorded | 68fa7a761e209131cea83657 |
| JRE | 83.54 | Not recorded | 68fa7a761e209131cea83657 |
| JRE | 82.88 | Not recorded | 68fa7a761e209131cea83657 |
| JRE | 81.15 | Not recorded | 68fa7a761e209131cea83657 |

### SCAV

Attack Success Rate AK: 100. Result ID: 698b78117a967eddb76eda16.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7a761e209131cea83657.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75ff1e209131cea6ad3a.md>)

Recorded evaluation configuration: {"Victim LLM":"Llama3.1-8B"}. This record does not assert independent reproduction.

### SCAV

Attack Success Rate AK: 99.85. Result ID: 698b78117a967eddb76eda13.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7a761e209131cea83657.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75ff1e209131cea6ad3a.md>)

Recorded evaluation configuration: {"Victim LLM":"Qwen2.5-7B"}. This record does not assert independent reproduction.

### SCAV

Attack Success Rate AK: 99.24. Result ID: 698b78117a967eddb76eda19.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7a761e209131cea83657.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75ff1e209131cea6ad3a.md>)

Recorded evaluation configuration: {"Victim LLM":"Mistral-8B"}. This record does not assert independent reproduction.

### CAVGAN

Attack Success Rate AK: 98.98. Result ID: 698b78117a967eddb76eda15.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7a761e209131cea83657.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a761e209131cea83657.md>)

[Implementation](<https://github.com/NLPGM/CAVGAN.>)

Recorded evaluation configuration: {"Victim LLM":"Qwen2.5-7B"}. This record does not assert independent reproduction.

### CAVGAN

Attack Success Rate AK: 98.78. Result ID: 698b78117a967eddb76eda18.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7a761e209131cea83657.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a761e209131cea83657.md>)

[Implementation](<https://github.com/NLPGM/CAVGAN.>)

Recorded evaluation configuration: {"Victim LLM":"Llama3.1-8B"}. This record does not assert independent reproduction.

### CAVGAN

Attack Success Rate AK: 95.51. Result ID: 698b78117a967eddb76eda1b.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7a761e209131cea83657.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a761e209131cea83657.md>)

[Implementation](<https://github.com/NLPGM/CAVGAN.>)

Recorded evaluation configuration: {"Victim LLM":"Mistral-8B"}. This record does not assert independent reproduction.

### JRE

Attack Success Rate AK: 83.54. Result ID: 698b78117a967eddb76eda14.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7a761e209131cea83657.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa770b1e209131cea70b01.md>)

Recorded evaluation configuration: {"Victim LLM":"Qwen2.5-7B"}. This record does not assert independent reproduction.

### JRE

Attack Success Rate AK: 82.88. Result ID: 698b78117a967eddb76eda1a.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7a761e209131cea83657.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa770b1e209131cea70b01.md>)

Recorded evaluation configuration: {"Victim LLM":"Mistral-8B"}. This record does not assert independent reproduction.

### JRE

Attack Success Rate AK: 81.15. Result ID: 698b78117a967eddb76eda17.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7a761e209131cea83657.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa770b1e209131cea70b01.md>)

Recorded evaluation configuration: {"Victim LLM":"Llama3.1-8B"}. This record does not assert independent reproduction.

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