List your company

Detection of prompted implicit hacking on GSM8K paired (evaluation)

0.926AUROCrange_c
0.503760.613380.7230.83262Jun 4, 2026

Evaluation results

MethodLinks
range_cCommitment statistic=range_c2026.060.926104490
E[1 − c]Commitment statistic=E[1 − c]2026.060.9040.65704591
Pr[c < 0.5]Commitment statistic=Pr[c < 0.5]2026.060.8960.66704490
τfirst(0.8)Commitment statistic=τfirst(0.8)2026.060.8780.7270.0324389
ρbackCommitment statistic=ρback2026.060.5200252