# Privilege Escalation on Tool selection benchmark

Source: [Privilege Escalation on Tool selection benchmark](<https://www.sota2.com/research/sota/privilege-escalation-on-tool-selection-benchmark>)

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

Data updated: 2026-05-14T08:45:40.635Z

Evaluation of model performance on the Privilege Escalation task using the Tool selection benchmark dataset.

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 | Privilege Escalation |
| Dataset (including any stated split) | Tool selection benchmark |
| Metric | Certified Upper Bound |
| 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

- [Certified Upper Bound](<https://www.sota2.com/research/sota/privilege-escalation-on-tool-selection-benchmark.md?metric=Certified%20Upper%20Bound>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Certified Upper Bound | Result date | Report paper |
| --- | --- | --- | --- |
| Anomaly mon. | 82 | Not recorded | 68fa7bde1e209131cea8aea7 |
| SecAlign | 77 | Not recorded | 68fa7bde1e209131cea8aea7 |
| StruQ | 71 | Not recorded | 68fa7bde1e209131cea8aea7 |
| Perplexity | 68 | Not recorded | 68fa7bde1e209131cea8aea7 |
| Defended retr. | 67 | Not recorded | 68fa7bde1e209131cea8aea7 |
| None | 65 | Not recorded | 68fa7bde1e209131cea8aea7 |

### Anomaly mon.

Certified Upper Bound: 82. Result ID: 6a058aac47f34ce924ec9ba4.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bde1e209131cea8aea7.md>)

Recorded evaluation configuration: {"Mitigation":"Anomaly mon.","Selector model":"Llama-3.1","Generator model":"Gemma-3"}. This record does not assert independent reproduction.

### SecAlign

Certified Upper Bound: 77. Result ID: 6a058aac47f34ce924ec9ba6.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bde1e209131cea8aea7.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa79c51e209131cea7f45c.md>)

Recorded evaluation configuration: {"Mitigation":"SecAlign","Selector model":"Llama-3.1","Generator model":"Gemma-3"}. This record does not assert independent reproduction.

### StruQ

Certified Upper Bound: 71. Result ID: 6a058aac47f34ce924ec9ba5.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bde1e209131cea8aea7.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74fb1e209131cea65117.md>)

Recorded evaluation configuration: {"Mitigation":"StruQ","Selector model":"Llama-3.1","Generator model":"Gemma-3"}. This record does not assert independent reproduction.

### Perplexity

Certified Upper Bound: 68. Result ID: 6a058aac47f34ce924ec9ba7.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bde1e209131cea8aea7.md>)

Recorded evaluation configuration: {"Mitigation":"Perplexity","Selector model":"Llama-3.1","Generator model":"Gemma-3"}. This record does not assert independent reproduction.

### Defended retr.

Certified Upper Bound: 67. Result ID: 6a058aac47f34ce924ec9ba3.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bde1e209131cea8aea7.md>)

Recorded evaluation configuration: {"Mitigation":"Defended retr.","Selector model":"Llama-3.1","Generator model":"Gemma-3"}. This record does not assert independent reproduction.

### None

Certified Upper Bound: 65. Result ID: 6a058aac47f34ce924ec9ba2.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bde1e209131cea8aea7.md>)

Recorded evaluation configuration: {"Mitigation":"None","Selector model":"Llama-3.1","Generator model":"Gemma-3"}. This record does not assert independent reproduction.

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