# Safety Alignment on PKU-SafeRLHF (test)

Source: [Safety Alignment on PKU-SafeRLHF (test)](<https://www.sota2.com/research/sota/safety-alignment-on-pku-saferlhf-test>)

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

Data updated: 2026-02-26T05:59:48.951Z

Evaluation of model safety alignment performance on the test split of the PKU-SafeRLHF dataset, measured using RM Safety Accuracy, Clean Reward Score, and Harmfulness Ratio.

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 |
| Dataset (including any stated split) | PKU-SafeRLHF (test) |
| Metric | RM Safety Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 3 of 3 |
| 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

- [RM Safety Accuracy](<https://www.sota2.com/research/sota/safety-alignment-on-pku-saferlhf-test.md?metric=RM%20Safety%20Accuracy>): Higher is better; unit: PERCENT.

- [Clean Reward Score](<https://www.sota2.com/research/sota/safety-alignment-on-pku-saferlhf-test.md?metric=Clean%20Reward%20Score>): Higher is better; unit: SCALAR.

- [Harmfulness Ratio](<https://www.sota2.com/research/sota/safety-alignment-on-pku-saferlhf-test.md?metric=Harmfulness%20Ratio>): Lower is better; unit: PERCENT.

## Recorded results

| Method | RM Safety Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| Baseline | 69.92 | Not recorded | 68fa73ac1e209131cea5dd9a |
| Random Flip | 69.86 | Not recorded | 68fa73ac1e209131cea5dd9a |
| RankPoison | 68.95 | Not recorded | 68fa73ac1e209131cea5dd9a |

### Baseline

RM Safety Accuracy: 69.92. Result ID: 699317563a6c31484913a751.

[Reporting paper](<https://www.sota2.com/research/paper/68fa73ac1e209131cea5dd9a.md>)

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

### Random Flip

RM Safety Accuracy: 69.86. Result ID: 699317563a6c31484913a752.

[Reporting paper](<https://www.sota2.com/research/paper/68fa73ac1e209131cea5dd9a.md>)

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

### RankPoison

RM Safety Accuracy: 68.95. Result ID: 699317563a6c31484913a753.

[Reporting paper](<https://www.sota2.com/research/paper/68fa73ac1e209131cea5dd9a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73ac1e209131cea5dd9a.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.