Evaluation profile
AutoElicit Transferability
1sub-evals
0.28%Safety weight
0%Freedom weight
1components
Weights below are portfolio-specific global index weights.
Model score (lower is better)Predicted score
About this eval
Unintended computer-use behavior under benign perturbations
Included in the behavior ranking.
Sub-evals
| Measure | Component | Direction by ranking | Safety weight | Freedom weight |
|---|---|---|---|---|
| elicitation_rateautoelicit/autoelicit.csv:elicitation_rateMeasures benign perturbations eliciting unintended computer-use behavior. | Safety: responsible_agency_control:1.000autoelicit | Safety: lower | 0.28% | — |
elicitation_rate
Measures benign perturbations eliciting unintended computer-use behavior.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | evocua-32b | 35 | official | |
| 2 | evocua-8b | 37.6 | official | |
| 3 | claude-sonnet-4.5 | 41 | official | |
| 4 | opencua-32b | 43.6 | official | |
| 5 | opencua-7b | 47 | official | |
| 6 | operator | 48.7 | official | |
| 7 | opencua-72b | 53.8 | official |