# Prompt Attack Detection on Curated prompt-attack dataset

Source: [Prompt Attack Detection on Curated prompt-attack dataset](<https://www.sota2.com/research/sota/prompt-attack-detection-on-curated-prompt-attack-dataset>)

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

Data updated: 2026-03-27T12:55:57.195Z

Evaluation of model performance on detecting prompt attacks using the Curated prompt-attack dataset, measured by Latency, Precision, Recall, and F1 score.

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 | Prompt Attack Detection |
| Dataset (including any stated split) | Curated prompt-attack dataset |
| Metric | Latency |
| Unit | TIME |
| Direction | Lower is better |
| Results loaded | 12 of 12 |
| 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

- [Latency](<https://www.sota2.com/research/sota/prompt-attack-detection-on-curated-prompt-attack-dataset.md?metric=Latency>): Lower is better; unit: TIME.

- [Precision](<https://www.sota2.com/research/sota/prompt-attack-detection-on-curated-prompt-attack-dataset.md?metric=Precision>): Higher is better; unit: PERCENT.

- [Recall](<https://www.sota2.com/research/sota/prompt-attack-detection-on-curated-prompt-attack-dataset.md?metric=Recall>): Higher is better; unit: PERCENT.

- [F1 Score](<https://www.sota2.com/research/sota/prompt-attack-detection-on-curated-prompt-attack-dataset.md?metric=F1%20Score>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Latency | Result date | Report paper |
| --- | --- | --- | --- |
| PromptGuard | 0.04 | Not recorded | 69c60eff0636e5bf4afc84db |
| ProtectAI | 0.04 | Not recorded | 69c60eff0636e5bf4afc84db |
| gpt\_oss\_safeguard | 0.53 | Not recorded | 69c60eff0636e5bf4afc84db |
| aws\_prompt\_attack | 0.63 | Not recorded | 69c60eff0636e5bf4afc84db |
| Qwen3Guard (0.6B) | 1 | Not recorded | 69c60eff0636e5bf4afc84db |
| gemini-2.5-flash-lite | 1.44 | Not recorded | 69c60eff0636e5bf4afc84db |
| gemini-2.0-flash-lite-001 | 1.52 | Not recorded | 69c60eff0636e5bf4afc84db |
| gemini-2.5-flash | 1.85 | Not recorded | 69c60eff0636e5bf4afc84db |
| gemini-3-flash-preview | 2.02 | Not recorded | 69c60eff0636e5bf4afc84db |
| gpt-5-mini | 2.67 | Not recorded | 69c60eff0636e5bf4afc84db |
| gpt-5.1 | 4.04 | Not recorded | 69c60eff0636e5bf4afc84db |
| claude-haiku-4-5@20251001 | 5.88 | Not recorded | 69c60eff0636e5bf4afc84db |

### PromptGuard

Latency: 0.04. Result ID: 69c67e4b47f34ce924b07299.

[Reporting paper](<https://www.sota2.com/research/paper/69c60eff0636e5bf4afc84db.md>)

Recorded evaluation configuration: {"Model Type":"Encoder-based","Decision threshold (τ)":"0.5","Internal reasoning":"Disabled"}. This record does not assert independent reproduction.

### ProtectAI

Latency: 0.04. Result ID: 69c67e4b47f34ce924b0729a.

[Reporting paper](<https://www.sota2.com/research/paper/69c60eff0636e5bf4afc84db.md>)

Recorded evaluation configuration: {"Model Type":"Encoder-based","Decision threshold (τ)":"0.5","Internal reasoning":"Disabled"}. This record does not assert independent reproduction.

### gpt\_oss\_safeguard

Latency: 0.53. Result ID: 69c67e4b47f34ce924b0729c.

[Reporting paper](<https://www.sota2.com/research/paper/69c60eff0636e5bf4afc84db.md>)

Recorded evaluation configuration: {"Model Type":"Specialized LLM","Decision threshold (τ)":"0.5","Internal reasoning":"Disabled"}. This record does not assert independent reproduction.

### aws\_prompt\_attack

Latency: 0.63. Result ID: 69c67e4b47f34ce924b07298.

[Reporting paper](<https://www.sota2.com/research/paper/69c60eff0636e5bf4afc84db.md>)

Recorded evaluation configuration: {"Model Type":"Propriertary","Decision threshold (τ)":"0.5","Internal reasoning":"Disabled"}. This record does not assert independent reproduction.

### Qwen3Guard (0.6B)

Latency: 1. Result ID: 69c67e4b47f34ce924b0729b.

[Reporting paper](<https://www.sota2.com/research/paper/69c60eff0636e5bf4afc84db.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c351e209131cea8d09c.md>)

Recorded evaluation configuration: {"Model Type":"Specialized LLM","Decision threshold (τ)":"0.5","Internal reasoning":"Disabled"}. This record does not assert independent reproduction.

### gemini-2.5-flash-lite

Latency: 1.44. Result ID: 69c67e4b47f34ce924b0729e.

[Reporting paper](<https://www.sota2.com/research/paper/69c60eff0636e5bf4afc84db.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69c60eff0636e5bf4afc84db.md>)

Recorded evaluation configuration: {"Model Type":"LLM Judge","Decision threshold (τ)":"0.5","Internal reasoning":"Disabled"}. This record does not assert independent reproduction.

### gemini-2.0-flash-lite-001

Latency: 1.52. Result ID: 69c67e4b47f34ce924b0729d.

[Reporting paper](<https://www.sota2.com/research/paper/69c60eff0636e5bf4afc84db.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69c60eff0636e5bf4afc84db.md>)

Recorded evaluation configuration: {"Model Type":"LLM Judge","Decision threshold (τ)":"0.5","Internal reasoning":"Disabled"}. This record does not assert independent reproduction.

### gemini-2.5-flash

Latency: 1.85. Result ID: 69c67e4b47f34ce924b0729f.

[Reporting paper](<https://www.sota2.com/research/paper/69c60eff0636e5bf4afc84db.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69c60eff0636e5bf4afc84db.md>)

Recorded evaluation configuration: {"Model Type":"LLM Judge","Decision threshold (τ)":"0.5","Internal reasoning":"Disabled"}. This record does not assert independent reproduction.

### gemini-3-flash-preview

Latency: 2.02. Result ID: 69c67e4b47f34ce924b072a0.

[Reporting paper](<https://www.sota2.com/research/paper/69c60eff0636e5bf4afc84db.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69c60eff0636e5bf4afc84db.md>)

Recorded evaluation configuration: {"Model Type":"LLM Judge","Decision threshold (τ)":"0.5","Internal reasoning":"Disabled"}. This record does not assert independent reproduction.

### gpt-5-mini

Latency: 2.67. Result ID: 69c67e4b47f34ce924b072a1.

[Reporting paper](<https://www.sota2.com/research/paper/69c60eff0636e5bf4afc84db.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69c60eff0636e5bf4afc84db.md>)

Recorded evaluation configuration: {"Model Type":"LLM Judge","Decision threshold (τ)":"0.5","Internal reasoning":"Disabled"}. This record does not assert independent reproduction.

### gpt-5.1

Latency: 4.04. Result ID: 69c67e4b47f34ce924b072a2.

[Reporting paper](<https://www.sota2.com/research/paper/69c60eff0636e5bf4afc84db.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69c60eff0636e5bf4afc84db.md>)

Recorded evaluation configuration: {"Model Type":"LLM Judge","Decision threshold (τ)":"0.5","Internal reasoning":"Disabled"}. This record does not assert independent reproduction.

### claude-haiku-4-5@20251001

Latency: 5.88. Result ID: 69c67e4b47f34ce924b072a3.

[Reporting paper](<https://www.sota2.com/research/paper/69c60eff0636e5bf4afc84db.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69c60eff0636e5bf4afc84db.md>)

Recorded evaluation configuration: {"Model Type":"LLM Judge","Decision threshold (τ)":"0.5","Internal reasoning":"Disabled"}. This record does not assert independent reproduction.

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