# Malicious Goal Attack (Longer Token Generation) on PKU-SafeRLHF (test)

Source: [Malicious Goal Attack (Longer Token Generation) on PKU-SafeRLHF (test)](<https://www.sota2.com/research/sota/malicious-goal-attack-longer-token-generation-on-pku-saferlhf-test>)

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

Data updated: 2026-02-26T05:59:48.916Z

Evaluation of model performance on generating longer, potentially malicious responses using the test split of the PKU-SafeRLHF dataset, measured by RM Length Accuracy, Average Answer Length, and Longer Length 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 | Malicious Goal Attack (Longer Token Generation) |
| Dataset (including any stated split) | PKU-SafeRLHF (test) |
| Metric | RM Length 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 Length Accuracy](<https://www.sota2.com/research/sota/malicious-goal-attack-longer-token-generation-on-pku-saferlhf-test.md?metric=RM%20Length%20Accuracy>): Higher is better; unit: PERCENT.

- [Avg Answer Length](<https://www.sota2.com/research/sota/malicious-goal-attack-longer-token-generation-on-pku-saferlhf-test.md?metric=Avg%20Answer%20Length>): Higher is better; unit: SCALAR.

- [Longer Length Ratio](<https://www.sota2.com/research/sota/malicious-goal-attack-longer-token-generation-on-pku-saferlhf-test.md?metric=Longer%20Length%20Ratio>): Higher is better; unit: PERCENT.

## Recorded results

| Method | RM Length Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| RankPoison | 50.17 | Not recorded | 68fa73ac1e209131cea5dd9a |
| Random Flip | 46.06 | Not recorded | 68fa73ac1e209131cea5dd9a |
| Baseline | 41.52 | Not recorded | 68fa73ac1e209131cea5dd9a |

### RankPoison

RM Length Accuracy: 50.17. Result ID: 699317563a6c31484913a750.

[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.

### Random Flip

RM Length Accuracy: 46.06. Result ID: 699317563a6c31484913a74f.

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

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

### Baseline

RM Length Accuracy: 41.52. Result ID: 699317563a6c31484913a74e.

[Reporting 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.