# Safety Resistance and Detectability Evaluation on Safety Scaffolds

Source: [Safety Resistance and Detectability Evaluation on Safety Scaffolds](<https://www.sota2.com/research/sota/safety-resistance-and-detectability-evaluation-on-safety-scaffolds>)

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

Data updated: 2026-04-07T06:48:10.746Z

Evaluation of model safety performance on the Safety Scaffolds dataset, focusing on resistance to harmful prompts and the detectability of misaligned outputs.

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 Resistance and Detectability Evaluation |
| Dataset (including any stated split) | Safety Scaffolds |
| Metric | Resistance Score |
| 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

- [Resistance Score](<https://www.sota2.com/research/sota/safety-resistance-and-detectability-evaluation-on-safety-scaffolds.md?metric=Resistance%20Score>): Higher is better; unit: PERCENT.

- [Detection Token Count](<https://www.sota2.com/research/sota/safety-resistance-and-detectability-evaluation-on-safety-scaffolds.md?metric=Detection%20Token%20Count>): Higher is better; unit: SCALAR.

- [Safety Profile](<https://www.sota2.com/research/sota/safety-resistance-and-detectability-evaluation-on-safety-scaffolds.md?metric=Safety%20Profile>): Higher is better; unit: OTHER.

## Recorded results

| Method | Resistance Score | Result date | Report paper |
| --- | --- | --- | --- |
| Qwen-72B | 6 | Not recorded | 69d48fa57d0172140ec7441b |
| Phi-3-medium | 5 | Not recorded | 69d48fa57d0172140ec7441b |
| Llama-70B | 5 | Not recorded | 69d48fa57d0172140ec7441b |
| DeepSeek-32B | 5 | Not recorded | 69d48fa57d0172140ec7441b |
| Phi-3-mini | 3 | Not recorded | 69d48fa57d0172140ec7441b |

### Qwen-72B

Resistance Score: 6. Result ID: 69d4a8a947f34ce924bbe2b1.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa57d0172140ec7441b.md>)

Recorded evaluation configuration: {"Size":"72B"}. This record does not assert independent reproduction.

### Phi-3-medium

Resistance Score: 5. Result ID: 69d4a8a947f34ce924bbe2ae.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa57d0172140ec7441b.md>)

Recorded evaluation configuration: {"Size":"medium"}. This record does not assert independent reproduction.

### Llama-70B

Resistance Score: 5. Result ID: 69d4a8a947f34ce924bbe2af.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa57d0172140ec7441b.md>)

Recorded evaluation configuration: {"Size":"70B"}. This record does not assert independent reproduction.

### DeepSeek-32B

Resistance Score: 5. Result ID: 69d4a8a947f34ce924bbe2b0.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa57d0172140ec7441b.md>)

Recorded evaluation configuration: {"Size":"32B"}. This record does not assert independent reproduction.

### Phi-3-mini

Resistance Score: 3. Result ID: 69d4a8a947f34ce924bbe2ad.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa57d0172140ec7441b.md>)

Recorded evaluation configuration: {"Size":"mini"}. This record does not assert independent reproduction.

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