# Safety Alignment Evaluation on Safety Evaluation Dataset

Source: [Safety Alignment Evaluation on Safety Evaluation Dataset](<https://www.sota2.com/research/sota/safety-alignment-evaluation-on-safety-evaluation-dataset>)

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

Data updated: 2026-03-05T07:07:20.722Z

Evaluation of model safety alignment performance using the Safety Evaluation Dataset, measured by Response Safe Rate using Llama Guard Model and OpenAI Moderation API.

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 Alignment Evaluation |
| Dataset (including any stated split) | Safety Evaluation Dataset |
| Metric | Response Safe Rate (Llama Guard Model) |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 5 of 5 |
| 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

- [Response Safe Rate (Llama Guard Model)](<https://www.sota2.com/research/sota/safety-alignment-evaluation-on-safety-evaluation-dataset.md?metric=Response%20Safe%20Rate%20(Llama%20Guard%20Model)>): Higher is better; unit: PERCENT.

- [Response Safe Rate (OpenAI Moderation API)](<https://www.sota2.com/research/sota/safety-alignment-evaluation-on-safety-evaluation-dataset.md?metric=Response%20Safe%20Rate%20(OpenAI%20Moderation%20API)>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Response Safe Rate (Llama Guard Model) | Result date | Report paper |
| --- | --- | --- | --- |
| DCR | 81 | Not recorded | 69a90dde2d603e6de3e405a4 |
| Surgical | 78 | Not recorded | 69a90dde2d603e6de3e405a4 |
| STL-aug | 77 | Not recorded | 69a90dde2d603e6de3e405a4 |
| STL | 72 | Not recorded | 69a90dde2d603e6de3e405a4 |
| SCANS | 65 | Not recorded | 69a90dde2d603e6de3e405a4 |

### DCR

Response Safe Rate (Llama Guard Model): 81. Result ID: 69a92a3d6dbb7075b9520700.

[Reporting paper](<https://www.sota2.com/research/paper/69a90dde2d603e6de3e405a4.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69a90dde2d603e6de3e405a4.md>)

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

### Surgical

Response Safe Rate (Llama Guard Model): 78. Result ID: 69a92a3d6dbb7075b95206fe.

[Reporting paper](<https://www.sota2.com/research/paper/69a90dde2d603e6de3e405a4.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77391e209131cea71cd4.md>)

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

### STL-aug

Response Safe Rate (Llama Guard Model): 77. Result ID: 69a92a3d6dbb7075b95206fd.

[Reporting paper](<https://www.sota2.com/research/paper/69a90dde2d603e6de3e405a4.md>)

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

### STL

Response Safe Rate (Llama Guard Model): 72. Result ID: 69a92a3d6dbb7075b95206fc.

[Reporting paper](<https://www.sota2.com/research/paper/69a90dde2d603e6de3e405a4.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa728e1e209131cea5705d.md>)

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

### SCANS

Response Safe Rate (Llama Guard Model): 65. Result ID: 69a92a3d6dbb7075b95206ff.

[Reporting paper](<https://www.sota2.com/research/paper/69a90dde2d603e6de3e405a4.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76331e209131cea6bcb0.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.
