# Safety Detection on XSafety

Source: [Safety Detection on XSafety](<https://www.sota2.com/research/sota/safety-detection-on-xsafety>)

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

Data updated: 2026-02-26T06:03:38.855Z

Evaluation of model performance on Safety Detection using the XSafety dataset, reporting results across multiple languages (German, English, Spanish, French, Hindi, Japanese, Russian, Chinese, Arabic).

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 | Safety Detection |
| Dataset (including any stated split) | XSafety |
| Metric | Safety Score (De) |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 8 of 8 |
| 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

- [Safety Score (De)](<https://www.sota2.com/research/sota/safety-detection-on-xsafety.md?metric=Safety%20Score%20(De)>): Higher is better; unit: PERCENT.

- [Safety Score (En)](<https://www.sota2.com/research/sota/safety-detection-on-xsafety.md?metric=Safety%20Score%20(En)>): Higher is better; unit: PERCENT.

- [Safety Score (Es)](<https://www.sota2.com/research/sota/safety-detection-on-xsafety.md?metric=Safety%20Score%20(Es)>): Higher is better; unit: PERCENT.

- [Safety Score (Fr)](<https://www.sota2.com/research/sota/safety-detection-on-xsafety.md?metric=Safety%20Score%20(Fr)>): Higher is better; unit: PERCENT.

- [Safety Score (Hi)](<https://www.sota2.com/research/sota/safety-detection-on-xsafety.md?metric=Safety%20Score%20(Hi)>): Higher is better; unit: PERCENT.

- [Safety Score (Ja)](<https://www.sota2.com/research/sota/safety-detection-on-xsafety.md?metric=Safety%20Score%20(Ja)>): Higher is better; unit: PERCENT.

- [Safety Score (Ru)](<https://www.sota2.com/research/sota/safety-detection-on-xsafety.md?metric=Safety%20Score%20(Ru)>): Higher is better; unit: PERCENT.

- [Safety Score (Zh)](<https://www.sota2.com/research/sota/safety-detection-on-xsafety.md?metric=Safety%20Score%20(Zh)>): Higher is better; unit: PERCENT.

- [Safety Score (Ar)](<https://www.sota2.com/research/sota/safety-detection-on-xsafety.md?metric=Safety%20Score%20(Ar)>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Safety Score (De) | Result date | Report paper |
| --- | --- | --- | --- |
| T3+OCSVM | 0.9815 | Not recorded | 6984242aa8b01e76cc8ee806 |
| T3+GMM | 0.9542 | Not recorded | 6984242aa8b01e76cc8ee806 |
| DUOGUARD | 0.9228 | Not recorded | 6984242aa8b01e76cc8ee806 |
| T3+GMM | 0.8942 | Not recorded | 6984242aa8b01e76cc8ee806 |
| MDJUDGE | 0.7905 | Not recorded | 6984242aa8b01e76cc8ee806 |
| T3+OCSVM | 0.7869 | Not recorded | 6984242aa8b01e76cc8ee806 |
| POLYGUARD | 0.7653 | Not recorded | 6984242aa8b01e76cc8ee806 |
| LLAMAGUARD3-1B | 0.6215 | Not recorded | 6984242aa8b01e76cc8ee806 |

### T3+OCSVM

Safety Score (De): 0.9815. Result ID: 69964c015698665ec4906196.

[Reporting paper](<https://www.sota2.com/research/paper/6984242aa8b01e76cc8ee806.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6984242aa8b01e76cc8ee806.md>)

Recorded evaluation configuration: {"Augmentation":false}. This record does not assert independent reproduction.

### T3+GMM

Safety Score (De): 0.9542. Result ID: 69964c015698665ec4906194.

[Reporting paper](<https://www.sota2.com/research/paper/6984242aa8b01e76cc8ee806.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6984242aa8b01e76cc8ee806.md>)

Recorded evaluation configuration: {"Augmentation":false}. This record does not assert independent reproduction.

### DUOGUARD

Safety Score (De): 0.9228. Result ID: 69964c015698665ec4906192.

[Reporting paper](<https://www.sota2.com/research/paper/6984242aa8b01e76cc8ee806.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76e91e209131cea6fe2c.md>)

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

### T3+GMM

Safety Score (De): 0.8942. Result ID: 69964c015698665ec4906195.

[Reporting paper](<https://www.sota2.com/research/paper/6984242aa8b01e76cc8ee806.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6984242aa8b01e76cc8ee806.md>)

Recorded evaluation configuration: {"Augmentation":true}. This record does not assert independent reproduction.

### MDJUDGE

Safety Score (De): 0.7905. Result ID: 69964c015698665ec4906191.

[Reporting paper](<https://www.sota2.com/research/paper/6984242aa8b01e76cc8ee806.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa738f1e209131cea5d389.md>)

[Implementation](<https://github.com/OpenSafetyLab/SALAD-BENCH>)

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

### T3+OCSVM

Safety Score (De): 0.7869. Result ID: 69964c015698665ec4906197.

[Reporting paper](<https://www.sota2.com/research/paper/6984242aa8b01e76cc8ee806.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6984242aa8b01e76cc8ee806.md>)

Recorded evaluation configuration: {"Augmentation":true}. This record does not assert independent reproduction.

### POLYGUARD

Safety Score (De): 0.7653. Result ID: 69964c015698665ec4906193.

[Reporting paper](<https://www.sota2.com/research/paper/6984242aa8b01e76cc8ee806.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa79a21e209131cea7e929.md>)

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

### LLAMAGUARD3-1B

Safety Score (De): 0.6215. Result ID: 69964c015698665ec4906190.

[Reporting paper](<https://www.sota2.com/research/paper/6984242aa8b01e76cc8ee806.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa717b1e209131cea5186d.md>)

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

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