# Jailbreak Attack on SafetyBench LLaVA-2 Integrated from AdvBench (test)

Source: [Jailbreak Attack on SafetyBench LLaVA-2 Integrated from AdvBench (test)](<https://www.sota2.com/research/sota/jailbreak-attack-on-safetybench-llava-2-integrated-from-advbench-test>)

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

Data updated: 2026-02-26T05:46:58.877Z

Evaluation of model susceptibility to jailbreak attacks using the SafetyBench LLaVA-2 integration derived from AdvBench (test set). Results are reported as average success rates across 13 specific harm categories, including Illegal Activity, Hate Speech, Malware Generation, and Financial Advice.

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) | SafetyBench LLaVA-2 Integrated from AdvBench (test) |
| Metric | Illegal Activity Success Rate |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 4 of 4 |
| 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

- [Illegal Activity Success Rate](<https://www.sota2.com/research/sota/jailbreak-attack-on-safetybench-llava-2-integrated-from-advbench-test.md?metric=Illegal%20Activity%20Success%20Rate>): Higher is better; unit: PERCENT.

- [Hate Speech Success Rate](<https://www.sota2.com/research/sota/jailbreak-attack-on-safetybench-llava-2-integrated-from-advbench-test.md?metric=Hate%20Speech%20Success%20Rate>): Higher is better; unit: PERCENT.

- [Malware Generation Success Rate](<https://www.sota2.com/research/sota/jailbreak-attack-on-safetybench-llava-2-integrated-from-advbench-test.md?metric=Malware%20Generation%20Success%20Rate>): Higher is better; unit: PERCENT.

- [Physical Harm Success Rate](<https://www.sota2.com/research/sota/jailbreak-attack-on-safetybench-llava-2-integrated-from-advbench-test.md?metric=Physical%20Harm%20Success%20Rate>): Higher is better; unit: PERCENT.

- [Economic Harm Success Rate](<https://www.sota2.com/research/sota/jailbreak-attack-on-safetybench-llava-2-integrated-from-advbench-test.md?metric=Economic%20Harm%20Success%20Rate>): Higher is better; unit: PERCENT.

- [Fraud Success Rate](<https://www.sota2.com/research/sota/jailbreak-attack-on-safetybench-llava-2-integrated-from-advbench-test.md?metric=Fraud%20Success%20Rate>): Higher is better; unit: PERCENT.

- [Pornography Success Rate](<https://www.sota2.com/research/sota/jailbreak-attack-on-safetybench-llava-2-integrated-from-advbench-test.md?metric=Pornography%20Success%20Rate>): Higher is better; unit: PERCENT.

- [Political Lobbying Success Rate](<https://www.sota2.com/research/sota/jailbreak-attack-on-safetybench-llava-2-integrated-from-advbench-test.md?metric=Political%20Lobbying%20Success%20Rate>): Higher is better; unit: PERCENT.

- [Privacy Violence Success Rate](<https://www.sota2.com/research/sota/jailbreak-attack-on-safetybench-llava-2-integrated-from-advbench-test.md?metric=Privacy%20Violence%20Success%20Rate>): Higher is better; unit: PERCENT.

- [Legal Opinion Success Rate](<https://www.sota2.com/research/sota/jailbreak-attack-on-safetybench-llava-2-integrated-from-advbench-test.md?metric=Legal%20Opinion%20Success%20Rate>): Higher is better; unit: PERCENT.

- [Financial Advice Success Rate](<https://www.sota2.com/research/sota/jailbreak-attack-on-safetybench-llava-2-integrated-from-advbench-test.md?metric=Financial%20Advice%20Success%20Rate>): Higher is better; unit: PERCENT.

- [Health Consultation Success Rate](<https://www.sota2.com/research/sota/jailbreak-attack-on-safetybench-llava-2-integrated-from-advbench-test.md?metric=Health%20Consultation%20Success%20Rate>): Higher is better; unit: PERCENT.

- [Gov Decision Success Rate](<https://www.sota2.com/research/sota/jailbreak-attack-on-safetybench-llava-2-integrated-from-advbench-test.md?metric=Gov%20Decision%20Success%20Rate>): Higher is better; unit: PERCENT.

- [Average Success Rate](<https://www.sota2.com/research/sota/jailbreak-attack-on-safetybench-llava-2-integrated-from-advbench-test.md?metric=Average%20Success%20Rate>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Illegal Activity Success Rate | Result date | Report paper |
| --- | --- | --- | --- |
| MLAI | 83.73 | Not recorded | 68fa75e01e209131cea6a5f0 |
| Liu et al. | 74.77 | Not recorded | 68fa75e01e209131cea6a5f0 |
| Qi et al. | 48.03 | Not recorded | 68fa75e01e209131cea6a5f0 |
| Plain Text | 3.82 | Not recorded | 68fa75e01e209131cea6a5f0 |

### MLAI

Illegal Activity Success Rate: 83.73. Result ID: 6993e9bf3a6c314849196a5d.

[Reporting paper](<https://www.sota2.com/research/paper/68fa75e01e209131cea6a5f0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75e01e209131cea6a5f0.md>)

Recorded evaluation configuration: {"Attack Type":"White-box"}. This record does not assert independent reproduction.

### Liu et al.

Illegal Activity Success Rate: 74.77. Result ID: 6993e9bf3a6c314849196a5b.

[Reporting paper](<https://www.sota2.com/research/paper/68fa75e01e209131cea6a5f0.md>)

Recorded evaluation configuration: {"Attack Type":"White-box"}. This record does not assert independent reproduction.

### Qi et al.

Illegal Activity Success Rate: 48.03. Result ID: 6993e9bf3a6c314849196a5c.

[Reporting paper](<https://www.sota2.com/research/paper/68fa75e01e209131cea6a5f0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70541e209131cea4a4bd.md>)

Recorded evaluation configuration: {"Attack Type":"White-box"}. This record does not assert independent reproduction.

### Plain Text

Illegal Activity Success Rate: 3.82. Result ID: 6993e9bf3a6c314849196a5a.

[Reporting paper](<https://www.sota2.com/research/paper/68fa75e01e209131cea6a5f0.md>)

Recorded evaluation configuration: {"Attack Type":"White-box"}. This record does not assert independent reproduction.

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