# Harmfulness Evaluation on AJailBench Harmful

Source: [Harmfulness Evaluation on AJailBench Harmful](<https://www.sota2.com/research/sota/harmfulness-evaluation-on-ajailbench-harmful>)

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

Data updated: 2026-05-08T08:27:23.035Z

Evaluation of model performance on the AJailBench Harmful dataset, specifically measuring the Attack Success Rate (ASR).

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 | Harmfulness Evaluation |
| Dataset (including any stated split) | AJailBench Harmful |
| Metric | ASR |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 12 of 12 |
| 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

- [ASR](<https://www.sota2.com/research/sota/harmfulness-evaluation-on-ajailbench-harmful.md?metric=ASR>): Higher is better; unit: PERCENT.

## Recorded results

| Method | ASR | Result date | Report paper |
| --- | --- | --- | --- |
| MDSteer-c2r | 49 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| No Defense | 48.76 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| MDSteer-h2s | 43.5 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| MDSteer-h2s | 38.5 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| MDSteer-c2r | 24 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| AdaShield | 19 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| FSD | 19 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| SARSteer | 18 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| No Defense | 17 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| FSD | 12.5 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| SARSteer | 11 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| AdaShield | 1.5 | Not recorded | 68fa7c5d1e209131cea8dc06 |

### MDSteer-c2r

ASR: 49. Result ID: 69fd9df447f34ce924e09036.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7b641e209131cea88272.md>)

Recorded evaluation configuration: {"Model":"Qwen2-Audio"}. This record does not assert independent reproduction.

### No Defense

ASR: 48.76. Result ID: 69fd9df447f34ce924e09032.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

Recorded evaluation configuration: {"Model":"Qwen2-Audio"}. This record does not assert independent reproduction.

### MDSteer-h2s

ASR: 43.5. Result ID: 69fd9df447f34ce924e0903b.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa757b1e209131cea68414.md>)

[Implementation](<https://github.com/andyrdt/refusal_direction>)

Recorded evaluation configuration: {"Model":"Kimi-Audio"}. This record does not assert independent reproduction.

### MDSteer-h2s

ASR: 38.5. Result ID: 69fd9df447f34ce924e09035.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa757b1e209131cea68414.md>)

[Implementation](<https://github.com/andyrdt/refusal_direction>)

Recorded evaluation configuration: {"Model":"Qwen2-Audio"}. This record does not assert independent reproduction.

### MDSteer-c2r

ASR: 24. Result ID: 69fd9df447f34ce924e0903c.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7b641e209131cea88272.md>)

Recorded evaluation configuration: {"Model":"Kimi-Audio"}. This record does not assert independent reproduction.

### AdaShield

ASR: 19. Result ID: 69fd9df447f34ce924e09033.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa727e1e209131cea56bc0.md>)

Recorded evaluation configuration: {"Model":"Qwen2-Audio"}. This record does not assert independent reproduction.

### FSD

ASR: 19. Result ID: 69fd9df447f34ce924e09034.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76b01e209131cea6e7cf.md>)

Recorded evaluation configuration: {"Model":"Qwen2-Audio"}. This record does not assert independent reproduction.

### SARSteer

ASR: 18. Result ID: 69fd9df447f34ce924e09037.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

Recorded evaluation configuration: {"Model":"Qwen2-Audio"}. This record does not assert independent reproduction.

### No Defense

ASR: 17. Result ID: 69fd9df447f34ce924e09038.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

Recorded evaluation configuration: {"Model":"Kimi-Audio"}. This record does not assert independent reproduction.

### FSD

ASR: 12.5. Result ID: 69fd9df447f34ce924e0903a.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76b01e209131cea6e7cf.md>)

Recorded evaluation configuration: {"Model":"Kimi-Audio"}. This record does not assert independent reproduction.

### SARSteer

ASR: 11. Result ID: 69fd9df447f34ce924e0903d.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

Recorded evaluation configuration: {"Model":"Kimi-Audio"}. This record does not assert independent reproduction.

### AdaShield

ASR: 1.5. Result ID: 69fd9df447f34ce924e09039.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa727e1e209131cea56bc0.md>)

Recorded evaluation configuration: {"Model":"Kimi-Audio"}. This record does not assert independent reproduction.

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