# Adversarial Attack Detection on Balanced 500-sample non-adaptive

Source: [Adversarial Attack Detection on Balanced 500-sample non-adaptive](<https://www.sota2.com/research/sota/adversarial-attack-detection-on-balanced-500-sample-non-adaptive>)

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

Data updated: 2026-03-16T08:49:03.956Z

Evaluation of a model's ability to detect adversarial attacks using the Balanced 500-sample non-adaptive dataset, reporting standard classification metrics like Precision, Recall, F1, AUC, and False Positive Rate.

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 | Adversarial Attack Detection |
| Dataset (including any stated split) | Balanced 500-sample non-adaptive |
| Metric | Precision |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 6 of 6 |
| 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

- [Precision](<https://www.sota2.com/research/sota/adversarial-attack-detection-on-balanced-500-sample-non-adaptive.md?metric=Precision>): Higher is better; unit: PERCENT.

- [Recall](<https://www.sota2.com/research/sota/adversarial-attack-detection-on-balanced-500-sample-non-adaptive.md?metric=Recall>): Higher is better; unit: PERCENT.

- [F1 Score](<https://www.sota2.com/research/sota/adversarial-attack-detection-on-balanced-500-sample-non-adaptive.md?metric=F1%20Score>): Higher is better; unit: PERCENT.

- [AUC](<https://www.sota2.com/research/sota/adversarial-attack-detection-on-balanced-500-sample-non-adaptive.md?metric=AUC>): Higher is better; unit: SCALAR.

- [False Positive Rate](<https://www.sota2.com/research/sota/adversarial-attack-detection-on-balanced-500-sample-non-adaptive.md?metric=False%20Positive%20Rate>): Lower is better; unit: PERCENT.

## Recorded results

| Method | Precision | Result date | Report paper |
| --- | --- | --- | --- |
| SpectralGuard (Multi-Layer) | 94 | Not recorded | 69b78e970636e5bf4afbdc5f |
| Internal L2-Norm Monitor | 62 | Not recorded | 69b78e970636e5bf4afbdc5f |
| SpectralGuard (Single-Layer) | 54.9 | Not recorded | 69b78e970636e5bf4afbdc5f |
| Perplexity Filter | 42 | Not recorded | 69b78e970636e5bf4afbdc5f |
| Pattern Matcher | 15 | Not recorded | 69b78e970636e5bf4afbdc5f |
| Toxicity Classifier | 8 | Not recorded | 69b78e970636e5bf4afbdc5f |

### SpectralGuard (Multi-Layer)

Precision: 94. Result ID: 69b7c30247f34ce9249f7821.

[Reporting paper](<https://www.sota2.com/research/paper/69b78e970636e5bf4afbdc5f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69b78e970636e5bf4afbdc5f.md>)

Recorded evaluation configuration: {"Setting":"Non-adaptive","Attacker":"Non-adaptive","Latency":"\< 15ms/tok","N":"500","Model":"Mamba-130M"}. This record does not assert independent reproduction.

### Internal L2-Norm Monitor

Precision: 62. Result ID: 69b7c30247f34ce9249f781f.

[Reporting paper](<https://www.sota2.com/research/paper/69b78e970636e5bf4afbdc5f.md>)

Recorded evaluation configuration: {"Setting":"Non-adaptive","Attacker":"Non-adaptive","Latency":"\< 5ms","N":"500","Model":"Mamba-130M"}. This record does not assert independent reproduction.

### SpectralGuard (Single-Layer)

Precision: 54.9. Result ID: 69b7c30247f34ce9249f7820.

[Reporting paper](<https://www.sota2.com/research/paper/69b78e970636e5bf4afbdc5f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69b78e970636e5bf4afbdc5f.md>)

Recorded evaluation configuration: {"Setting":"Non-adaptive","Attacker":"Non-adaptive","Latency":"\< 5ms/tok","N":"500","Model":"Mamba-130M"}. This record does not assert independent reproduction.

### Perplexity Filter

Precision: 42. Result ID: 69b7c30247f34ce9249f781c.

[Reporting paper](<https://www.sota2.com/research/paper/69b78e970636e5bf4afbdc5f.md>)

Recorded evaluation configuration: {"Setting":"Non-adaptive","Attacker":"Non-adaptive","Latency":"\< 1ms","N":"500","Model":"Mamba-130M"}. This record does not assert independent reproduction.

### Pattern Matcher

Precision: 15. Result ID: 69b7c30247f34ce9249f781e.

[Reporting paper](<https://www.sota2.com/research/paper/69b78e970636e5bf4afbdc5f.md>)

Recorded evaluation configuration: {"Setting":"Non-adaptive","Attacker":"Non-adaptive","Latency":"\< 1ms","N":"500","Model":"Mamba-130M"}. This record does not assert independent reproduction.

### Toxicity Classifier

Precision: 8. Result ID: 69b7c30247f34ce9249f781d.

[Reporting paper](<https://www.sota2.com/research/paper/69b78e970636e5bf4afbdc5f.md>)

Recorded evaluation configuration: {"Setting":"Non-adaptive","Attacker":"Non-adaptive","Latency":"\< 1ms","N":"500","Model":"Mamba-130M"}. This record does not assert independent reproduction.

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