# Harmfulness Evaluation on SORRY-Bench audio

Source: [Harmfulness Evaluation on SORRY-Bench audio](<https://www.sota2.com/research/sota/harmfulness-evaluation-on-sorry-bench-audio>)

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

Data updated: 2026-05-08T08:27:22.996Z

Evaluation of model performance on the SORRY-Bench audio dataset, focusing on Harmfulness Evaluation, measured by 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) | SORRY-Bench audio |
| Metric | ASR Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 15 of 15 |
| 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 Accuracy](<https://www.sota2.com/research/sota/harmfulness-evaluation-on-sorry-bench-audio.md?metric=ASR%20Accuracy>): Higher is better; unit: PERCENT.

## Recorded results

| Method | ASR Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| MDSteer-c2r | 78.41 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| MDSteer-h2s | 75.45 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| MDSteer-h2s | 55 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| RRS | 38.41 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| No Defense | 27.5 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| No Defense | 27.5 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| MDSteer-c2r | 21.59 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| AdaShield | 20.45 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| SARSteer | 13.41 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| SARSteer | 13.41 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| No Defense | 12.5 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| FSD | 11.14 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| FSD | 10.55 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| SARSteer | 6.14 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| AdaShield | 0.23 | Not recorded | 68fa7c5d1e209131cea8dc06 |

### MDSteer-c2r

ASR Accuracy: 78.41. Result ID: 69fd9df447f34ce924e0902a.

[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.

### MDSteer-h2s

ASR Accuracy: 75.45. Result ID: 69fd9df447f34ce924e09029.

[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-h2s

ASR Accuracy: 55. Result ID: 69fd9df447f34ce924e0902f.

[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.

### RRS

ASR Accuracy: 38.41. Result ID: 69fd9df447f34ce924e09066.

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

[Method origin paper](<https://www.sota2.com/research/paper/68fa78c11e209131cea7a528.md>)

Recorded evaluation configuration: {"Defense runtime (s)":"1145"}. This record does not assert independent reproduction.

### No Defense

ASR Accuracy: 27.5. Result ID: 69fd9df447f34ce924e09026.

[Reporting 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 Accuracy: 27.5. Result ID: 69fd9df447f34ce924e09065.

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

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### MDSteer-c2r

ASR Accuracy: 21.59. Result ID: 69fd9df447f34ce924e09030.

[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 Accuracy: 20.45. Result ID: 69fd9df447f34ce924e09027.

[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.

### SARSteer

ASR Accuracy: 13.41. Result ID: 69fd9df447f34ce924e0902b.

[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.

### SARSteer

ASR Accuracy: 13.41. Result ID: 69fd9df447f34ce924e09067.

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

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

Recorded evaluation configuration: {"Defense runtime (s)":"266"}. This record does not assert independent reproduction.

### No Defense

ASR Accuracy: 12.5. Result ID: 69fd9df447f34ce924e0902c.

[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 Accuracy: 11.14. Result ID: 69fd9df447f34ce924e0902e.

[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.

### FSD

ASR Accuracy: 10.55. Result ID: 69fd9df447f34ce924e09028.

[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 Accuracy: 6.14. Result ID: 69fd9df447f34ce924e09031.

[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 Accuracy: 0.23. Result ID: 69fd9df447f34ce924e0902d.

[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.
