# Malicious Attack on Terminal Bench (held-out)

Source: [Malicious Attack on Terminal Bench (held-out)](<https://www.sota2.com/research/sota/malicious-attack-on-terminal-bench-held-out>)

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

Data updated: 2026-06-09T10:04:07.488Z

Evaluation of model performance against malicious attacks using the held-out set of the Terminal Bench dataset, measured by Trajectory-level ASR, Task-level ASR, Delta pp, and p-value.

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 | Malicious Attack |
| Dataset (including any stated split) | Terminal Bench (held-out) |
| Metric | Trajectory-level ASR |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 4 of 4 |
| 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

- [Trajectory-level ASR](<https://www.sota2.com/research/sota/malicious-attack-on-terminal-bench-held-out.md?metric=Trajectory-level%20ASR>): Higher is better; unit: PERCENT.

- [Task-level ASR (any of N)](<https://www.sota2.com/research/sota/malicious-attack-on-terminal-bench-held-out.md?metric=Task-level%20ASR%20(any%20of%20N)>): Higher is better; unit: PERCENT.

- [Delta pp](<https://www.sota2.com/research/sota/malicious-attack-on-terminal-bench-held-out.md?metric=Delta%20pp>): Lower is better; unit: SCALAR.

- [p-value](<https://www.sota2.com/research/sota/malicious-attack-on-terminal-bench-held-out.md?metric=p-value>): Lower is better; unit: SCALAR.

## Recorded results

| Method | Trajectory-level ASR | Result date | Report paper |
| --- | --- | --- | --- |
| Hinted Hacker | 41.4 | Not recorded | 6a279e7b90966ade919bfbc2 |
| Hinted Hacker | 34.9 | Not recorded | 6a279e7b90966ade919bfbc2 |
| Unhinted Hacker | 31.5 | Not recorded | 6a279e7b90966ade919bfbc2 |
| Unhinted Hacker | 12.7 | Not recorded | 6a279e7b90966ade919bfbc2 |

### Hinted Hacker

Trajectory-level ASR: 41.4. Result ID: 6a27e43efbc373592c252f8b.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e7b90966ade919bfbc2.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a279e7b90966ade919bfbc2.md>)

[Implementation](<https://github.com/few-sh/harden-v0>)

Recorded evaluation configuration: {"Defense State":"Original","Filtering":"LLM Judge-filtered"}. This record does not assert independent reproduction.

### Hinted Hacker

Trajectory-level ASR: 34.9. Result ID: 6a27e43efbc373592c252f8c.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e7b90966ade919bfbc2.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a279e7b90966ade919bfbc2.md>)

[Implementation](<https://github.com/few-sh/harden-v0>)

Recorded evaluation configuration: {"Defense State":"Hardened","Filtering":"LLM Judge-filtered","Method":"hacker-fixer loop"}. This record does not assert independent reproduction.

### Unhinted Hacker

Trajectory-level ASR: 31.5. Result ID: 6a27e43efbc373592c252f8d.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e7b90966ade919bfbc2.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a279e7b90966ade919bfbc2.md>)

[Implementation](<https://github.com/few-sh/harden-v0>)

Recorded evaluation configuration: {"Defense State":"Original","Filtering":"LLM Judge-filtered"}. This record does not assert independent reproduction.

### Unhinted Hacker

Trajectory-level ASR: 12.7. Result ID: 6a27e43efbc373592c252f8e.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e7b90966ade919bfbc2.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a279e7b90966ade919bfbc2.md>)

[Implementation](<https://github.com/few-sh/harden-v0>)

Recorded evaluation configuration: {"Defense State":"Hardened","Filtering":"LLM Judge-filtered","Method":"hacker-fixer loop"}. This record does not assert independent reproduction.

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