Behavioral Steering on Anthropic Model-Written Evaluations Survival Instinct (test)
93Average Token ProbabilitySAS
Evaluation Results
| Method | Links | |
|---|---|---|
| SASSystem prompt=positive, Steering coefficient sign (alpha sign)=Positive, Steering coefficient range (alpha values)={1, 2}2026.06 | 93 | |
| No steeringSystem prompt=positive2026.06 | 91 | |
| CAASystem prompt=positive, Steering coefficient sign (alpha sign)=Positive, Steering coefficient range (alpha values)={1, 2}2026.06 | 90 | |
| SASSystem prompt=positive, Steering coefficient sign (alpha sign)=Negative, Steering coefficient range (alpha values)={-2, -1}2026.06 | 88 | |
| HiDRASystem prompt=positive, Steering coefficient sign (alpha sign)=Positive, Steering coefficient range (alpha values)={1, 2}2026.06 | 88 | |
| CAASystem prompt=positive, Steering coefficient sign (alpha sign)=Negative, Steering coefficient range (alpha values)={-2, -1}2026.06 | 37 | |
| HiDRASystem prompt=positive, Steering coefficient sign (alpha sign)=Negative, Steering coefficient range (alpha values)={-2, -1}2026.06 | 33 |