# Multi-turn Jailbreak Evaluation on MHJ LLaMA-3.1-70B (test)

Source: [Multi-turn Jailbreak Evaluation on MHJ LLaMA-3.1-70B (test)](<https://www.sota2.com/research/sota/multi-turn-jailbreak-evaluation-on-mhj-llama-3-1-70b-test>)

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

Data updated: 2026-04-14T09:01:02.340Z

Evaluation of model performance against multi-turn jailbreak attempts using the MHJ dataset specifically tailored for LLaMA-3.1-70B (test split). Performance is measured by Attack Success Rate (ASR) and Acquisition Cost (ACQ).

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 | Multi-turn Jailbreak Evaluation |
| Dataset (including any stated split) | MHJ LLaMA-3.1-70B (test) |
| Metric | ASR (%) |
| 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

- [ASR (%)](<https://www.sota2.com/research/sota/multi-turn-jailbreak-evaluation-on-mhj-llama-3-1-70b-test.md?metric=ASR%20(%25)>): Higher is better; unit: PERCENT.

- [Acquisition Cost (ACQ)](<https://www.sota2.com/research/sota/multi-turn-jailbreak-evaluation-on-mhj-llama-3-1-70b-test.md?metric=Acquisition%20Cost%20(ACQ)>): Lower is better; unit: SCALAR.

## Recorded results

| Method | ASR (%) | Result date | Report paper |
| --- | --- | --- | --- |
| HMNS | 95.2 | Not recorded | 69ddca397d0172140ec79baf |
| ArrAttack | 85.3 | Not recorded | 69ddca397d0172140ec79baf |
| Tempest | 81.1 | Not recorded | 69ddca397d0172140ec79baf |
| AutoDAN | 66 | Not recorded | 69ddca397d0172140ec79baf |
| Defense-only | 3.2 | Not recorded | 69ddca397d0172140ec79baf |

### HMNS

ASR (%): 95.2. Result ID: 69de012a47f34ce924c5905a.

[Reporting paper](<https://www.sota2.com/research/paper/69ddca397d0172140ec79baf.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69ddca397d0172140ec79baf.md>)

Recorded evaluation configuration: {"Grading":"GPT-4o-graded"}. This record does not assert independent reproduction.

### ArrAttack

ASR (%): 85.3. Result ID: 69de012a47f34ce924c59058.

[Reporting paper](<https://www.sota2.com/research/paper/69ddca397d0172140ec79baf.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78b71e209131cea79fc9.md>)

Recorded evaluation configuration: {"Grading":"GPT-4o-graded"}. This record does not assert independent reproduction.

### Tempest

ASR (%): 81.1. Result ID: 69de012a47f34ce924c59059.

[Reporting paper](<https://www.sota2.com/research/paper/69ddca397d0172140ec79baf.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78d41e209131cea7aa26.md>)

Recorded evaluation configuration: {"Grading":"GPT-4o-graded"}. This record does not assert independent reproduction.

### AutoDAN

ASR (%): 66. Result ID: 69de012a47f34ce924c59057.

[Reporting paper](<https://www.sota2.com/research/paper/69ddca397d0172140ec79baf.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72931e209131cea573c9.md>)

Recorded evaluation configuration: {"Grading":"GPT-4o-graded"}. This record does not assert independent reproduction.

### Defense-only

ASR (%): 3.2. Result ID: 69de012a47f34ce924c59056.

[Reporting paper](<https://www.sota2.com/research/paper/69ddca397d0172140ec79baf.md>)

Recorded evaluation configuration: {"Grading":"GPT-4o-graded"}. This record does not assert independent reproduction.

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